Packet Radio Networks

Packet Radio Networks are multiaccess networks in

which not all nodes can hear the transmission of all
other nodes, this feature is characteristic for radio
communication
We will focus on the effect of partial connectivity on the
multiaccess techniques rather than the physical
characteristics of the radio broadcast medium
The topology of a radio network can be described by a
graph, G = (N, L), where N is a set of nodes and L is a
set of links, each link correspond to an ordered pair of
nodes, (i, j), and indicates that transmission from i can
be heard at j

Information Networks – p.1/28

 Packet Radio Networks
In some situations node j might be able to hear node i
but i is unable to hear j , in such a case (i, j) ∈ L but
(j, i) 6∈ L

1 3 4

5 6

Information Networks – p.2/28

 Packet Radio Networks
Our assumption about communication in this
multiaccess medium is that if node i transmits a packet,
that packet will be correctly received by node j if and
only if
There is a link from i to j , i.e. (i, j) ∈ L, and
No other node k for which (k, j) ∈ L is transmitting
while i is transmitting, and
j itself is not transmitting while i is transmitting
A large number of links in a graph is not necessarily
desirable, it does increase the number of nodes that
can communicate directly but also increases the
likelihood of collision

Information Networks – p.3/28

 Packet Radio Networks
The question is now how much traffic can be carried in
such a network?
We define a collision-free set as a set of links that can
carry packets simultaneously with no collisions at the
receiving ends of the links
We can order the links and represent each collision-free
set as a vector of 0’s and 1’s called a collision-free
vector (CFV), where the l th component of a CFV is 1 if
and only if the lth link is in the corresponding
collision-free set

Information Networks – p.4/28

 Packet Radio Networks
Some CFVs for our example graph is

(1,2) (1,5) (2,1) (3,2) (3,4) (3,6) (3,5) (4,6) (5,3) (6,3) (6,4)

1 0 0 0 0 0 0 1 0 0 0
0 1 0 0 0 0 0 1 0 0 0
1 0 0 0 0 0 0 0 0 1 0
0 0 1 0 0 0 0 1 1 0 0
0 0 1 0 0 0 0 0 0 0 1
0 0 0 1 0 0 0 0 0 0 0
0 0 0 0 1 0 0 0 0 0 0
0 0 0 0 0 1 0 0 0 0 0

Information Networks – p.5/28

 TDM for Packet Radio Nets
Choose some collection of CFVs {xi } and cycle
between them by TDM, i.e. in the ith slot of a TDM cycle
all links with a 1 in xi can carry packets
There are no collisions and the fraction of time that a
given link can carry packets is the fraction of the CFVs
that contain a 1 in the position corresponding to that
link, so with J CFVs
1X
f= xi
J
i

is the fraction of time each link can be used

Information Networks – p.6/28

 TDM for Packet Radio Nets
By repeating a CFV xi a certain number of times in a
TDM frame, with αi the fraction of frame slots using xi
we get
X
f= αi x i
i

as the fractional utilization of each link

P P
A vector of the form i αi xi with i αi = 1 and αi ≥ 0 is
called a convex combination of the vectors {xi }
We can state the above result as; any convex
combination of CFVs can be approached arbitrarily
closely as a fractional link utilization vector through the
use of TDM

Information Networks – p.7/28

 TDM for Packet Radio Nets
Suppose we are using some sort of collision resolution
method in the network, at any given time, the vector of
links that are transmitting successfully is a CFV
By averaging this vector over time we get a vector
whose lth component is the fraction of time that the l th
link is carrying packets successfully
This is also a convex combination of CFVs, thus we see
that any link utilization that is achievable with collision
resolution is also achievable by TDM
One disadvantage of TDM is that the delays are longer
than necessary for a lightly loaded network

Information Networks – p.8/28

 TDM for Packet Radio Nets
However, if all nodes have only a small number of
incoming links, many nodes can transmit
simultaneously and the waiting time for TDM slot is
reduced
Another problem with the TDM approach is that the
nodes are usually mobile, and thus the topology of the
network is constantly changing, this means that the
CFVs keep changing, requiring frequent updates of the
TDM schedule
The problem of determining whether a potential vector
of link utilizations is a convex combination of a given set
of CFVs has a computational time that increases very
rapidly with the number of links in the network

Information Networks – p.9/28

 FDM for Packet Radio Nets
FDM can also be used for packet radio networks in a
similar way to TDM, all links in a CFV can use the same
frequency band simultaneously, so in principle the links
can carry the same amount of traffic as in TDM
This approach is used in cellular radio networks for
mobile voice communication
The area covered by the network is divided into a large
number of local areas called cells, each cell has a
number of frequency bands for use within that cell
The frequency bands used by one cell can be reused
by other cells that are sufficiently separated from each
other to avoid interference

Information Networks – p.10/28

Collision Resolution for Packet Radio Nets
One complication in packet radio nets is obtaining
feedback information, suppose that the links (3, 5) and
(4, 6) contain packets in a given slot, then node 6
perceives a collision and node 5 correctly receives a
packet
If nodes 5 and 6 send feedback information, node 3 will
experience a feedback collision
A second problem is that if a node perceives a collision,
it does not know if any of the packets were addressed
to it
Thus we cannot assume perfect (0, 1, e) feedback and
the splitting algorithms cannot be used and the
stabilization techniques require substantial revisions

Information Networks – p.11/28

Collision Resolution for Packet Radio Nets
Slotted and unslotted Aloha are still applicable, and to a
certain extent, some of the ideas of carrier sensing and
reservation can still be used
We start by analyzing how slotted Aloha works in this
case
When an unbacklogged node receives a packet to
transmit (either a new packet entering the network or a
packet in transit that needs to be forwarded to another
node), it transmits the packet in the next slot
If no acknowledgment (ack) of correct reception arrives
within some time-out period, the node becomes
backlogged and the packet is retransmitted after a
random delay

Information Networks – p.12/28

Slotted Aloha for Packet Radio Nets
A backlogged node becomes unbacklogged when all its
packets have been transmitted and acked successfully
The simplest way to return acks to the transmitting node
is that if i sends a packet to j that must be forwarded on
to some other node k , then if i hears j ’s transmission to
k that serves as an ack of the (i, j) transmission
This however needs to be complemented with some
way to ack packets from i that are destined for j
Further, if j successfully relays the packet to k but i fails
to hear this due to a collision, an unnecessary
retransmission from i to j is done and j need to ack this
retransmission in some other way since j has already
forwarded the packet to k

Information Networks – p.13/28

Slotted Aloha for Packet Radio Nets
Another approach is for each node to include explicit
acks for the last few packets it has received in each
outgoing packet
This requires a node to send a dummy packet carrying
ack information if the node has no data to send for
some period
A third approach is to provide time at the end of each
slot for explicit acks of packets received within the slot
We will now analyze what happens in slotted Aloha for
a heavily loaded network

Information Networks – p.14/28

Slotted Aloha for Packet Radio Nets
Assume that all nodes are backlogged all the time and
has packets to send on all outgoing links at all times
We assume that the nodes have infinite buffers to store
the backlogged packets
For all nodes i and j , let qij be the probability that node
i transmits a packet to node j in any given slot
P
Let Qi = j qij be the probability that node i transmits
to any node
We let qij = 0 if (i, j) 6∈ L

Information Networks – p.15/28

Slotted Aloha for Packet Radio Nets
Let pij be the probability that a transmission on (i, j) is
successful
Under our assumption of heavy loading each node
transmits or not in a slot independently of all other
nodes
Since pij is the probability that none of the other nodes
that can reach j , including j itself, is transmitting we get
Y
pij = (1 − Qj ) (1 − Qk )
k:(k,j)∈L,k6=i

Finally, the rate fij of successful transmissions per slot

on link (i, j) is fij = qij pij

Information Networks – p.16/28

Slotted Aloha for Packet Radio Nets
Given the attempt rates qij we can now compute the
link throughputs fij under the heavy-loading
assumptions, but we would rather be able to find the
attempt rates qij that will yield a desired set of
throughputs (if that set of throughputs is feasible)
This latter problem can be solved iteratively, given a
desired throughput fij , we start with an initial qij 0 = 0,

and forP each iteration n = 0, 1, 2, . . . we first compute

Qni = j qij n (which thus will all be 0 when n = 0) and
Q
then pij = (1 − Qj ) k:(k,j)6∈L,k6=i (1 − Qnk ) (which thus will
n n

all be 1 when n = 0), and then we get next iteration of

n+1
qij by qij = pfijn
ij

Information Networks – p.17/28

Slotted Aloha for Packet Radio Nets
Using this iterative procedure we get qij 1 ≥ q 0 , thus
ij
Q1i ≥ Q0i and p1ij ≤ p0ij , and qij
2 ≥ q 1 , and so on, as long
ij
as none of the Qni exceed 1
Thus as long as none of the Qni exceed 1 we get that qij
n

is nondecreasing and pnij is nonincreasing with

successive iterations n
It follows that either some Qni exceed 1 at some iteration
n approaches a limit, q ∗ , and in this limit, with
n or else qij ij
corresponding Q∗i and p∗ij we have a solution to our
equations that for the attempt rate qij ∗ gives the

throughput fij

Information Networks – p.18/28

Slotted Aloha for Packet Radio Nets
If any Qni > 1 for some n then there is no solution to our
equations and the given throughput fij is infeasible
If we know the input rates to the network, and the
routes over which the sessions flow, we can in principle
determine the steady-state rates fij0 at which the links
must handle traffic
We would like to choose the throughputs of each link
under heavy load to exceed these steady-state rates so
the backlogs do not build up indefinitely

Information Networks – p.19/28

Slotted Aloha for Packet Radio Nets
We can then search for the largest number β > 1 for
which fij = βfij0 is feasible under heavy-load
assumptions, given this largest fij , and the
corresponding attempt rates qij , we can empty out the
backlog as it develops
One problem here is that if some nodes are backlogged
and others are not, the unbacklogged nodes no longer
choose their transmission times independently, so it is
possible in some odd cases that some backlogged
nodes build up more backlog when other nodes are
unbacklogged than they do when all nodes are
backlogged

Information Networks – p.20/28

Slotted Aloha for Packet Radio Nets
One way to avoid this difficulty is for new packets at a
node to join the backlog immediately rather than being
able to transmit in next slot, this increases delay under
light-loading conditions
The other way is to hope for the best, to some extent
one has to do this anyway since with a changing
topology one cannot maintain carefully controlled
attempt rates
One reason for focusing on the heavily loaded case is
that the number of links entering each node is usually
small so the attempt rates can be moderately high even
under heavy-loading assumption

Information Networks – p.21/28

Slotted Aloha for Packet Radio Nets
The other reason is that stabilization is much harder
here since a node cannot help itself too much by
adjusting its own attempt rates, since other nodes may
cause congestion without experiencing congestion
themselves
So far, we have viewed the set of links as given,
however, if a node increases its transmitter power, its
transmission will reach a larger set of nodes, it is
however desirable to keep the power level relatively low
so that each node has a moderately small set of
incoming and outgoing links, somewhere around 8
could be good according to a theoretical analysis
(although with some questionable assumptions)

Information Networks – p.22/28

Carrier Sensing and Busy Tones
We have previously seen that carrier sensing yielded a
considerable improvement in the situation where all
nodes could here all other nodes and the propagation
delay is small
For line-of-sight radio, the propagation delay is typically
small relative to packet transmission times, so it’s
reasonable to explore how well it will work here
We have the hidden node problem, if node i is
transmitting to node j and node k also wants to transmit
to node j there is no guarantee that node k can hear i,
so carrier sensing can prevent some collisions but not
all

Information Networks – p.23/28

Carrier Sensing and Busy Tones
With carrier sensing we lose the uniform slotting
structure and thus lose some of the advantage that
slotted Aloha has over unslotted Aloha
Also, radio transmission is subject to fading and
variable noise so detecting another transmitting node is
hard to do in a short time
For all these reasons carrier sensing is not very
effective for packet radio
One approach to improving the performance of carrier
sensing is to use a busy tone, whenever any node
detects a packet being transmitted, it starts to send a
signal, called a busy tone, in a separate frequency band

Information Networks – p.24/28

Carrier Sensing and Busy Tones
When node i starts to send a packet to node j , then
node j (along with all other nodes that can hear node i)
will start to send a busy tone
All the nodes that can hear j will thus avoid
transmitting, and assuming that the nodes that can hear
j is the same as then nodes j can hear it follows that j
will experience no collision
A problem is that when node i starts to send a packet,
all nodes in range of i will send busy tones, and thus
every node within range of any node in range of i will be
inhibited from transmitting

Information Networks – p.25/28

Carrier Sensing and Busy Tones
Assuming transmission radius of R, when node i starts
to transmit most nodes within radius of 2R of i will be
inhibited, this is typically about 4 times the number of
nodes within radius R from the receiving node, which is
the set of nodes that should be inhibited, thus from
throughput standpoint this is not very promising
Another variation is for a node to send busy tone only
after it receives the address part of the packet and
recognizes itself as the intended recipient, this has
besides more complexity also the disadvantage of
increasing the time β over which another node could
start transmission before hearing the busy tone

Information Networks – p.26/28

 Packet Radio Networks
In summary, for packet radio many more questions than
answers exist, both in terms of desirable structure and
how to analyze
In addition questions about modulation and detection
make the situation even more complex
It is often desirable to use spread-spectrum techniques
for sending packets, one of the consequences of this is
that if two packets are being received at once, the
receiver can often lock on to one with the other acting
only as wideband noise

Information Networks – p.27/28

 Packet Radio Networks
If different spread-spectrum codes are used for each
receiver, the receiver can look for only its own sequence
and thus reject simultaneous packets sent to other
receivers, but unwanted packets can arrive with much
higher power levels than desired packets and still cause
a collision

Information Networks – p.28/28