Ad Hoc Networks 169 (2025) 103750

Contents lists available at ScienceDirect

Ad Hoc Networks
journal homepage: www.elsevier.com/locate/adhoc

Capacity and delay analysis of multi-hop wireless networks

Jan Laarhuis a ,∗,1 , Alessandro Chiumento b ,2
a
Thales Nederland B.V., Zuidelijke Havenweg 40, 7554 RR, Hengelo, The Netherlands
b
University Twente, Drienerloweg 1, 7554 RR, Enschede, The Netherlands

ARTICLE INFO ABSTRACT

Keywords: The capacity and delay of wireless multi-hop networks are key performance indicators. They are needed in
MANET the design of such networks, and are also useful to assess how well applications can run over such networks.
Multi-hop However, finding analytical expressions for capacity and delay for a wireless multi-hop network with an
TDMA
arbitrary topology is hard. In this work, a two-step approach is followed to derive expressions for the maximum
Scheduling
achievable capacity and the minimum achievable delay in wireless multi-hop networks. These expressions are
Capacity
Delay
valid for fixed, conflict-free time-slot scheduling among the nodes, and when each nodes of the network has
always packets to send to each of its neighbours. In the first step, expressions for the capacity and delay of
the elementary topologies ’string’ and ’star’ are derived, and in the second step, these results are combined
to derive the capacity and delay values for a network with an arbitrary topology. This two-step approach is
applied to two types of wireless multi-hop networks: those whose nodes have an omni-directional antenna, and
those whose nodes have an electronically steerable directional (beam-steering) antenna. Using this approach,
we find that the capacity of a path on the network is not a decreasing function of the total number of nodes in
the network, as mostly found in literature, but rather a decreasing function of the number of neighbours of the
nodes on the path. The results show that the derived maximum value for the capacity (and minimum value
for the delay) in networks with beam-steering antennas is larger (lower) than that for networks with omni-
directional antennas as more efficient scheduling is possible for the former. The derived analytic expressions for
capacity and delay are valuable for the relative comparison of the performance of wireless multi-hop networks.

1. Introduction on nodes with a beam-steering antenna, that can point their narrow
beam in any direction of the azimuth. Such networks, for which we use
Wireless multi-hop networks exist for a long time and are used in the adjective ’beam-steering’, necessarily must use space–time division,
various applications. Examples are static mesh networks to provide i.e., STDMA. For both classical and beam-steering networks, we assume
Internet access in rural areas, and dynamic mobile ad-hoc networks that3 :
(MANET) that connect unmanned vehicles maneuvering in a maritime
area to fulfil a task. To assess the suitability of a wireless network 1. all nodes use their radio at the same frequency;
for such applications, or to compare different network concepts with 2. a node can only reach its direct neighbours using its radio;
regard to those applications, an estimation of their performance is 3. the capacity of all radio channels in the network is 𝑟;
needed. 4. time is divided in slots;
A well-known class of wireless multi-hop networks is based on 5. static, conflict-free time-slot scheduling applies;
nodes with a single omnidirectional antenna. For such networks we 6. all nodes always have data to send to each of their neighbours.
use the adjective ‘classical’. Although most research on such networks Consider an 𝑛-hop path in a network of 𝑁 nodes and an arbitrary
is based on 802.11-based radios using CSMA-CA to access the radio topology. This 𝑛-hop path (𝑛 ≤ 𝑁 − 1) consists of 𝑛 + 1 nodes and 𝑛
channel, in this article the focus is on time-division access of the radio
links between adjacent nodes on this path. The red nodes in Fig. 1 form
channel (TDMA). Another class of wireless multi-hop networks is based
a 4-hop path in a wireless mesh network. The blue nodes are off-path

∗ Corresponding author.
E-mail addresses: [email protected] (J. Laarhuis), [email protected] (A. Chiumento).
1
Jan Laarhuis is currently a research fellow at the Pervasive Systems group of the University Twente.
2
Alessandro Chiumento is with the Pervasive Systems group of the University Twente.
3
Theoretically, assumptions 2 and 3 imply that all nodes must have the same inter-node distance. This is not true in practice but it greatly simplifies the
analysis.

https://doi.org/10.1016/j.adhoc.2024.103750
Received 29 May 2024; Received in revised form 20 November 2024; Accepted 27 December 2024
Available online 3 January 2025
1570-8705/© 2025 The Authors. Published by Elsevier B.V. This is an open access article under the CC BY license (http://creativecommons.org/licenses/by/4.0/).
J. Laarhuis and A. Chiumento Ad Hoc Networks 169 (2025) 103750

The remainder of the paper is structured as follows. A review

of related work is first given in Section 2. Our contribution starts
with definitions and a simple mathematical formalism in Section 3.
In Section 4, expressions for capacity and delay in ‘classical’ MANETs
with an arbitrary topology are derived. A similar approach is applied
in Section 5 for ’beam-steering’ MANETs. The obtained results are
discussed and evaluated regarding their applicability in Section 6.
Conclusions are given in Section 7.

2. Comparing our work to the literature

There is a large body of work published on the performance of

multi-hop wireless networks. To familiarize, a good starting point is [1],
in which an overview of work on capacity in multi-hop wireless net-
works of the past two decades is given. The authors described results on
static and mobile networks, on homogeneous and heterogeneous net-
Fig. 1. Wireless mesh network with a 4-hop path. works, and on nodes with omni-directional and directional antenna’s.
Their overview started with the work of Gupta and Kumar [2], which
have derived that, in a network with randomly placed nodes, the
neighbours of the nodes on the path. The capacity of this path is defined capacity of a node is in the order of 𝑟∕(𝑁 log 𝑁), where 𝑟 and 𝑁
as the maximum throughput between the ingress node and the egress are the capacity of the wireless channel and the number of nodes in
node over this path. For low values of load, offered at the ingress node the network, respectively. Other citations in [1] presented variations
and exiting the network at the egress node, the throughput at the egress on the results in [2]. For a network with nodes having only omni-
node is expected to keep pace with the load. When load increases, the directional antenna’s, the main conclusion was that the capacity of
a node decreases when number of nodes increases. For networks in
throughput levels off to the capacity, and delay increases or even losses
which nodes have directional antennas for transmission and reception,
occur.
an increase in capacity of 2𝜋 2 ∕𝛼 𝛽 with respect to omni-directional
The load consists of data packets generated by communicating users
antennas was obtained, where 𝛼 and 𝛽 are the gain of the transmitting
of the network, which are not shown in Fig. 1. These users can be
and receiving antenna, respectively.
applications running at hosts connected to the network, whose data
All works cited in [1] considered the capacity of a single node,
packets flow in corresponding sessions over the network. For these
whereas we consider the capacity of an 𝑛-hop path in the network. So
users, the performance of the path connecting them is of relevance.
the results are not directly comparable. But since the primary function
The performance of individual nodes on the path is of less value and
of a network is to transport data packets between nodes connected to
therefore not considered in this work.
it, we consider the capacity of a path in the network more useful than
This paper assumes time-based, conflict-free scheduling of time slots
that of a single node. The work we describe in this paper applies to the
between neighbouring nodes, and that each node always has data to be
static, homogeneous category in [1]. As we will see, unlike previous
transmitted to each of its neighbours. These are the assumptions under
published work, but under the conditions described in Section 1, our
which closed form expressions for the capacity and minimum delay of
results show that the capacity of a path does not decrease with increas-
a path in the network will be derived.
ing path length 𝑛, nor with an increase of the total number of nodes
While it is hard for an arbitrary topology to find an optimal TDMA 𝑁 in the network, but rather it decreases when the maximum degree
schedule, those schedules can be derived for well-known topologies 𝑚𝑚𝑎𝑥 on the path increases.
such as ‘string’ and ‘star’. Therefore, we propose a new approach in Besides the works cited in [1], the body of work on capacity of wire-
which the schedules of string and star topologies are combined to less, multi-hop networks is larger. In [3], back in 1985, the scheduling
arrive at analytical expressions for the capacity and delay in arbitrary problem in packet radio networks with fixed node positions was anal-
topologies. In the first step of the approach, a path on the network is ysed using graph theory. The authors used the concept of compatibility
considered a string and the capacity and delay of this string are derived matrix, which provides links that can transmit simultaneously without
for both ‘classical’ networks and ’beam-steering’ networks. The second collisions. On this matrix, maximum cliques were found to which one
step takes into account that the path is not a true string because one or more (not necessarily adjacent) time slots of a frame were allocated.
or more nodes on the path can have more than two neighbours. The It provided a graph theoretical analysis without an expression for the
impact of the presence of these additional neighbours on the schedule (maximum achievable) capacity of the network.
is reflected in the expressions for capacity and delay. The work in [4] resembles similarities with our approach, for exam-
The contributions of this paper are threefold: ple their ‘disk graph’ model resembles our assumption that only direct
neighbours can be connected by radio, and their protocol interference
• derivation of an approach that leads to analytical expressions for model resembles our contention area. But [4] assumed a different
the capacity and the minimum achievable delay of a path in an communication pattern (many-to-one in wireless sensor networks) than
arbitrary topology of a wireless multi-hop network, under the in our case (arbitrary path (one-to-one) through the network). Their
above-mentioned assumptions; situation was more complex because they considered trees instead of
• application of the approach to networks consisting of nodes strings. Again, no analytic expressions were given for the capacity.
with omni-directional antennas (classical networks) and networks In [5] TDMA scheduling in networks with nodes that have ‘multiple
consisting of nodes with steerable directional antennas (beam- reception capacity’ was considered. This means that a node 𝑖 in [5] can
steering networks); simultaneously receive packets collision-free on 𝑝𝑖 orthogonal channels.
• instead of the abstract mathematical approach taken by the ma- The multiple reception capacity closely resembles the potential ’multi-
jority of the earlier works on the subject, a novel, more tangible direction reception’ capability of beam-steering nodes we will discuss
approach based on concepts known in the networking world in Section 5, in which the channels are orthogonal in the direction
(string and star topologies) is taken. only. [5] stated that when 𝑝𝑖 is larger than its number of neighbours,

2
J. Laarhuis and A. Chiumento Ad Hoc Networks 169 (2025) 103750

an optimal schedule can be found. Despite this statement, in [5] no Table 1

Used variables and their meaning.
analytic expression of the capacity was derived, except one valid under
specific, restricted conditions. Variable Meaning

Another publication that considers TDMA scheduling with multiple 𝑁 number of nodes in the network
𝑛 number of hops of a path in the network
reception capability is [6]. While a fully connected network, i.e., a
𝑠 number of time slots in a frame
single-hop network, multi-hop forwarding was scheduled to make op- TxOp transmission opportunity
timal use of the multiple reception capability. In [6] expressions were SimTx number of simultaneous TxOps in a path
given from which the capacity can be derived, but because of the fully 𝑚𝑖 number of neighbours of node 𝑖
connected network, the situation in [6] is not comparable with our 𝑀 total number of TxOp in a network
𝜌 scheduling density
situation.
𝜆 scheduling rate
We observe that closed analytic expressions for the capacity of 𝐶 capacity of a path in the network
wireless, multi-hop networks with an arbitrary topology can hardly 𝐷 delay of a data unit on a path in the network
be found in the literature. The reason is the NL-completeness of the
scheduling problem (more specifically the maximum matching prob-
lem) for wireless multi-hop networks [7–9]. Another observation from
literature is that all cited works do not consider the delay over a path. The scheduling density of a path in a multi-hop network depends
The latter can be attributed to the focus in the consulted literature on on the contention area for transmission, i.e., how many hops further
the capacity of a single node of the network, for which delay has no can there be a TxOp at the same time? This is a characteristic of the
(practical) meaning. type of antenna and medium access control (MAC) mechanism used in
Compared to the existing body of knowledge, our results extend it in the network, as we will see in Section 4 and Section 5. Since 𝜌 is a
two ways. First, by using a different approach we derive, under specific fractional number, to find the integral number of simultaneous TxOps
assumptions, closed analytical expressions for capacity and minimum (SimTx) in one time slot on a path with 𝑛 hops, ceiling must be applied:
delay of a path in the network. Second, by introducing the generic
𝑆 𝑖𝑚𝑇 𝑥 = ⌈𝜌 ⋅ 𝑛⌉ (2)
concepts ‘scheduling density’ and ‘scheduling rate’, which make the
approach applicable to a wide range of network and node types. For
example, the approach may incorporate other antenna types, or other The capacity of a path depends on the transmission bit rate 𝑟 and
medium access control schemes. the fraction of the time it can be used to forward data units over the
We noticed a reduction in publications on the subject of capac- path. With Eq. (1) and Eq. (2), and the scheduling density, the capacity
ity and delay of wireless multi-hop networks in the past ten years becomes:
compared to the previous era. Regardless of the age of the topic, 𝑟 ⋅ 𝑆 𝑖𝑚𝑇 𝑥
𝐶= . (3)
this remained an open problem for which we are providing closed- 𝑀
form expressions for capacity and minimum achievable delay under the
conditions mentioned. This equation can be illustrated with a simple example. Consider a
network with a string topology for which 𝑁 = 𝑛 + 1. For this situation,
3. Developing the mathematical formalism Eq. (3) reduces to 𝐶 = 𝑟 ⋅ ⌈𝜌 ⋅ 𝑛⌉∕(2 ⋅ 𝑛) where 2 ⋅ 𝑛 = 𝑀 is the number
of TxOps in a string topology. The capacity is at most 𝑟∕2 when 𝜌 = 1,
In this section a model is described from which expressions for where the factor 2 accounts for the forward and backward direction
capacity and delay are derived. We start with the general multi-hop over the path.
network shown in Fig. 1. A path through this network is shown by the The capacity of a path in the network can also be derived using the
red, dashed arrow. It has an ingress node and an egress node. A node number of time slots 𝑠 in a frame. In this case the capacity becomes:
𝑖 has degree 𝑚𝑖 meaning it has 𝑚𝑖 (direct) neighbours. A summary of 𝑟
𝐶= , (4)
all variables introduced in this section and their meaning is given in 𝑠
Table 1. and by combining Eq. (3) and Eq. (4) the following relation is obtained:
Without loss of generality, a time slot is defined as the period in 𝑀
𝑠= . (5)
which one data unit (packet or frame) can be transmitted. A more 𝑆 𝑖𝑚𝑇 𝑥
specific definition for transmission is the following:
To analyse the delay in a multi-hop path, the concept of scheduling
Definition 1 (Transmit Opportunity TxOp). A transmit opportunity for rate is introduced:
node 𝑖 allows it to transmit to a neighbour in a time slot without
unrecoverable collisions. Definition 3 (Scheduling Rate 𝜆). The scheduling rate 𝜆 ∈ ⟨0, 1] in a
multi-hop path is the number of consecutive hops that can be taken in
A frame is a sequence of 𝑠 time slots in which each node of the
one time slot. It is a measure of the progression of a particular data unit
network has a TxOp for each of its neighbours. Node 𝑖 needs 𝑚𝑖 TxOps
through the path.
to transmit to each of its neighbours. The total required number of
TxOps 𝑀 in a network is the summation over all nodes: The scheduling rate is determined by the number of consecutive
hops taken and the number of time slots required for it. When one
∑
𝑁
𝑀= 𝑚𝑖 . (1) hop is taken in each time slot, 𝜆 = 1. When there are time slots in
𝑖=1 which no TxOp is scheduled, 𝜆 < 1. Similar to the scheduling density,
because 𝜆 is a fractional number, to find the integral number of time
Because multiple TxOps may occur simultaneously in a path in one slots needed to traverse a path with 𝑛 hops, ceiling must be applied.
time slot, 𝑠 can be less than 𝑀. To model this, the scheduling density With the scheduling rate, the delay of a data unit on an 𝑛-hop path in
in defined as follows: terms of time slots is calculated as:
𝑛
Definition 2 (Scheduling Density 𝜌). The scheduling density 𝜌 ∈ ⟨0, 1] in 𝐷 = ⌈ ⌉. (6)
𝜆
a multi-hop path is the number of transmit opportunities per hop in the
same time slot. It is a measure of the number of data units in transit on This equation can also be illustrated for the same string topology as
the path. used for the capacity. If 𝜆 = 1 then it takes 𝑛 time slots to travel the

3
J. Laarhuis and A. Chiumento Ad Hoc Networks 169 (2025) 103750

Table 2
Calculation of variable values for different path lengths 𝑛 for string topology and omni-
directional antenna.
𝑛 𝑀𝑟𝑒𝑞 1 2 𝑀𝑠𝑡𝑟𝑖𝑛𝑔 𝑠 𝐶̃ 𝐷𝑠𝑙𝑜𝑡
3 4 2 2 4 4 0,25 4;4
4 6 2 2 6 4 0,25 6;6
5 8 2 2 8 4 0,25 8;8
6 10 2 2 10 4 0,25 10;10
7 12 2 2 12 4 0,25 12;12
8 14 2 2 14 4 0,25 14;14
Fig. 2. Hidden node and exposed node situations for CSMA-CA and TDMA. 9 16 2 2 16 4 0,25 16;16
10 18 2 2 18 4 0,25 18;18
11 20 2 2 20 4 0,25 20;20
12 22 2 2 22 4 0,25 22;22
𝑛-hop string. When 𝜆 < 1 is smaller, the number of time slots required, 13 24 2 2 24 4 0,25 24;24
and thus the delay, will be larger. 14 26 2 2 26 4 0,25 26;26
The expressions for capacity and delay derived above are applicable 15 28 2 2 28 4 0,25 28;28
16 30 2 2 30 4 0,25 30;30
to arbitrary multi-hop networks, but the variables 𝜌 and 𝜆 cannot be 17 32 2 2 32 4 0,25 32;32
calculated for an arbitrary multi-hop network. However, for networks 18 34 2 2 34 4 0,25 34;34
with specific topologies, like string topology and star topology, they can 19 36 2 2 36 4 0,25 36;36
20 38 2 2 38 4 0,25 38;38
be derived. Each individual node in a network and its neighbours can be
considered a star topology. This is shown in Fig. 1 by the nodes within
the blue dashed circle. Each individual path between ingress and egress
node in a network can be considered a string topology, as illustrated by Lemma 1 (Occurrence ). The occurrence 𝑖 is the number of time slots
the red dashed arrow in Fig. 1. In the following sections this is used to in a frame with 𝑆 𝑖𝑚𝑇 𝑥𝑖 transmissions. Since two values of 𝑆 𝑖𝑚𝑇 𝑥 are
derive expressions for the capacity and delay of multi-hop networks possible, also two values of occurrences exist. The value of 𝑖 depends on
with arbitrary topology. the remainder of 𝑞1 = 𝜌𝑠𝑡𝑟𝑖𝑛𝑔 ⋅ 𝑛 and 𝑞2 = 𝜌𝑠𝑡𝑟𝑖𝑛𝑔 ⋅ (𝑛 − 1) as follows:
⎧ 2 for rem(𝑞 ) = 0
⎪ 1
4. Capacity analysis in ‘classical’ MANETs 1 = ⎨ 1 for rem(𝑞1 ) < rem(𝑞2 )
⎪ 2 for rem(𝑞1 ) ≥ rem(𝑞2 )
⎩
In this section the scheduling density 𝜌 and scheduling rate 𝜆 (8)
are derived for ‘classical’ MANETs, i.e., MANETs whose nodes have ⎧ 1 for rem(𝑞1 ) = 0
⎪
one omni-directional antenna. The string and star topologies are first 2 = ⎨ 2 for rem(𝑞1 ) < rem(𝑞2 )
considered in isolation in Section 4.1 and Section 4.2, and then they ⎪ 1 for rem(𝑞1 ) ≥ rem(𝑞2 )
⎩
are combined in Section 4.3 to make an estimate of the capacity and
delay of a network with an arbitrary topology.
Using this lemma, the number of time slots per frame in a string
with TDMA are calculated as the sum of the occurrences:
4.1. Analysis for string topology
𝑠𝑠𝑡𝑟𝑖𝑛𝑔 = 1 + 2 , (9)

As stated in Section 3 the scheduling density depends on the con- and the total number of TxOps in a frame (𝑀𝑠𝑡𝑟𝑖𝑛𝑔 ) is calculated using
tention area of a network. In CSMA-CA networks, the contention area Eq. (7) and Lemma 1:
is determined by the well-known ‘exposed node’ and ‘hidden node’ 𝑀𝑠𝑡𝑟𝑖𝑛𝑔 = 𝑆 𝑖𝑚𝑇 𝑥1 ⋅ 1 + 𝑆 𝑖𝑚𝑇 𝑥2 ⋅ 2 . (10)
problems (see Fig. 2). In the exposed node situation (upper left part),
T3 cannot transmit to R4 when T2 transmits at the same time to R1.
T3 could have sent to R4 but CSMA-CA operation prevents that. In the Table 1 shows the results of all variables mentioned above for dif-
hidden node situation (upper right part of Fig. 2), T3 cannot transmit to ferent lengths of the string 𝑛. The required number of TxOps in column
R2 when T1 is already transmitting to R2. Due to CSMA-CA operation, ’𝑀𝑟𝑒𝑞 ’ has been calculated using Eq. (1). The remaining columns show
T3 is not aware of this and a collision can occur (see, e.g., [10]). the results of the equations derived in this section. When the remainder
of 𝑞1 and 𝑞2 is either 0 or 0.5, which is the case for TDMA, then both
When scheduling transmissions in a TDMA frame, the hidden node
1 and 2 are 2. The column ‘𝑀𝑠𝑡𝑟𝑖𝑛𝑔 ’ shows the number of TxOps
and exposed node problems are avoided when the transmissions are
calculated using Eq. (10). As can be seen, all values correspond to the
scheduled as shown in the lower part of Fig. 2. For this situation,
required number of TxOps shown in column ‘𝑀𝑟𝑒𝑞 ’. Using Eq. (4) the
one transmission is scheduled on every two hops. Thus the scheduling ̃
capacity relative to the link bit rate 𝑟 is shown in column ’𝐶’.
density of a string with TDMA is 𝜌𝑠𝑡𝑟𝑖𝑛𝑔 = 0.5.
The delay of data packets over the string depends on the number of
The number of TxOps in a string depends on the scheduling density
time slots required for the propagation of the data packets through the
𝜌𝑠𝑡𝑟𝑖𝑛𝑔 , the number of hops 𝑛 of the path, and also whether the edge
string. To evenly distribute the delay for forward and backward traffic
node is involved in a scheduled transmission. An edge node must be over the string, transmissions with opposite directions are scheduled
left unscheduled when the schedule pattern dictates transmission to or in adjacent time slots in Fig. 3 in an example 11-hop string. The blue
reception from a imaginary neighbour beyond the edge node. In that coloured arrows show the forwarding from node 1 to node 12, and the
case, there is one hop less available for scheduling. Eq. (2) is used to green arrows the forwarding from node 12 to node 1. As can be seen in
calculate the value of SimTx for the two situations: this figure, in both directions every other time slot a consecutive hop
𝑆 𝑖𝑚𝑇 𝑥1 = ⌈𝜌𝑠𝑡𝑟𝑖𝑛𝑔 ⋅ 𝑛⌉; 𝑆 𝑖𝑚𝑇 𝑥2 = ⌈𝜌𝑠𝑡𝑟𝑖𝑛𝑔 ⋅ (𝑛 − 1)⌉. (7) can be taken. This means that for both directions, the scheduling rate
𝜆𝑠𝑡𝑟𝑖𝑛𝑔 ,𝑡𝑑 𝑚𝑎 = 0.5. Eq. (6) is used to calculate the delay in terms of time
slots, and is shown in column ‘𝐷𝑠𝑙𝑜𝑡 ’ of Table 2. In this column ‘a;b’
To derive an expression for the capacity, the following lemma is means that ‘a’ is the delay in one direction of the path, and ‘b’ is the
needed. delay in the other direction.

4
J. Laarhuis and A. Chiumento Ad Hoc Networks 169 (2025) 103750

Fig. 5. TDMA scheduling for an arbitrary path.

Fig. 3. TDMA scheduling in a string for various frames.

transmit to one of their respective non-string neighbours. Similarly, all
middle nodes on the path can simultaneously receive from one of their
respective non-string neighbours.
Like in Fig. 3, the red arrows in the matrix show the time slot
allocation for the string. The blue arrows show the additional transmis-
sions scheduled to send to and receive from the non-string neighbours.
This leads to additional time slots for this specific topology. In Fig. 5.a
the middle nodes all have one extra neighbour. In the corresponding
schedule matrix is shown that only two extra time slots are needed to
accommodate transmissions to and from these extra neighbours. When
Fig. 4. TDMA scheduling in a 5-node star. some of the middle nodes have four neighbours, Fig. 5.b. shows that
in this case four additional time slots are needed. The following lemma
generalizes this mechanism and presents an analytical expression for
4.2. Analysis for star topology the number of time slots for an arbitrary path.

The capacity and delay of a star topology is now analysed. An 𝑁- Lemma 2 (Time Slots for Arbitrary Path). If the maximum degree of the
node star is characterized by one hub and 𝑁 − 1 edge nodes. The middle nodes of a path is 𝑚𝑚𝑎𝑥 , then 𝛥𝑠 = 2 ⋅ (𝑚𝑚𝑎𝑥 − 2) extra time slots are
capacity of a star is defined by the maximum throughput between required in the time slot schedule.
any pair of edge nodes, i.e., from one edge node to the hub, and The total number of time slots in a TDMA frame for an arbitrary
from the hub to another edge node. For an 𝑁-node star there are
𝑛-node path with a maximum degree 𝑚𝑚𝑎𝑥 is now:
no simultaneous transmissions and hence 𝑠𝑠𝑡𝑎𝑟 = 2 ⋅ (𝑁 − 1) = 𝑀.
𝑠𝑝𝑎𝑡ℎ = 𝑠𝑠𝑡𝑟𝑖𝑛𝑔 + 𝛥𝑠𝑡𝑑 𝑚𝑎
A straightforward scheduling of one frame is shown in Fig. 4. First
(11)
all edge nodes are scheduled to transmit to the hub, and then the
= 4 + 2 ⋅ (𝑚𝑚𝑎𝑥 − 2),
hub is scheduled to transmit to each of the edge nodes. Between any
combination of edge nodes only one time slot per frame is available,
The capacity can be calculated using Eq. (11) and Eq. (4).
and the capacity of an 𝑁-node star is calculated using Eq. (4).
The delay for an arbitrary path can be obtained by incorporating
The delay is defined as the time needed for the transfer of data
the additional time slots in the scheduling rate for an arbitrary path
between any two edge nodes. The delay in a star topology in terms
as follows. From Definition 3 follows that the number of consecutive
of time slots depends on which edge nodes are involved in a schedule.
hops in one frame is 𝜆𝑠𝑡𝑟𝑖𝑛𝑔 ⋅ 𝑠𝑠𝑡𝑟𝑖𝑛𝑔 . This number does not change as a
For the schedule in Fig. 4, the delay from node 2 to node 5 is eight time
consequence of the added time slots. Thus the scheduling rate 𝜆𝑝𝑎𝑡ℎ for
slots while the delay from node 5 to node 2 is only two time slots. On
an arbitrary path can now be written as:
average, the delay in a star is 𝑁 − 1 time slots. In terms of frames, the 𝑠𝑠𝑡𝑟𝑖𝑛𝑔
delay is always one frame. 𝜆𝑝𝑎𝑡ℎ = 𝜆𝑠𝑡𝑟𝑖𝑛𝑔 ⋅ , (12)
𝑠𝑠𝑡𝑟𝑖𝑛𝑔 + 𝛥𝑠𝑡𝑑 𝑚𝑎

4.3. Analysis of an arbitrary topology and Eq. (6) can be used to calculate the delay of an arbitrary path.

In the previous subsections an 𝑛-hop path within the network has 5. Capacity analysis in beam-steering MANETs
been analysed as an 𝑛-hop string topology, and each node of the net-
work and its neighbours as a star topology. In this section the developed So far, the analysis of capacity and delay in MANETs is applied to
concepts are combined to derive an expression for the upper bound nodes with omni-directional antennas. In this section we extend the
of the capacity and the minimum achievable delay in an arbitrary developed idea to MANETs in which the nodes have a electronically
topology. The basic idea is to first consider a path as a pure string steerable directional antenna. The same structure as with ‘classical’
for which the derived expressions in Section 4.1 apply. Then the non- MANETs is followed. This means the string topology, star topology and
string nodes are taken into account in the scheduling. This results in arbitrary topology are described in separate subsections. This section
additional time slots to satisfy all required TxOps, and this in turn starts with a concise description of the properties of beam-steering
affects the capacity and delay of the path. antennas.
Fig. 5 illustrates this approach for a 4-hop string (red nodes) where
the middle nodes have more than two neighbours, where the non-string 5.1. Properties of beam-steering antennas
neighbours are shown as blue nodes. We assume that each non-string
neighbour of a node on the string and each non-string neighbour of an This section introduces beam-steering antennas. For the purpose of
adjacent node on the string satisfy the conditions for the contention this paper, a beam-steering antenna is a phased array of many radiating
area. This means that all middle nodes of the path can simultaneously elements that has the following characteristics and properties:

5
J. Laarhuis and A. Chiumento Ad Hoc Networks 169 (2025) 103750

1. each radiating element has its own analog–digital and digital–

analog converter;
2. the azimuth (360 degrees) is discretized into beams of width 𝛥𝜃
in direction 𝜃𝑖 = 𝑖 ⋅ 𝛥𝜃 , 𝑖 = 0, 1, .. 360
𝛥𝜃
;
3. the array can cover any direction 𝜃𝑖 of the azimuth;
4. the array can transmit:

(a) in one specific direction 𝜃𝑖 using a narrow, high-gain Fig. 6. Contention area for space–time multiplexing (STDMA) with beam steering
beam; nodes.

(b) in the entire azimuth using a low-gain (omni-directional)

beam; Table 3
Calculation of variable values for different path lengths 𝑛 for string topology and beam-
steering antenna.
5. the array can receive:
𝑛 𝑀𝑟𝑒𝑞 1 2 𝑀𝑠𝑡𝑟𝑖𝑛𝑔 𝑠 𝐶̃ 𝐷𝑠𝑙𝑜𝑡
(a) simultaneously and independently from all directions 𝜃𝑖 3 4 1 2 4 3 0,33 2;4
of the azimuth at high gain. This is called ‘multi-direction 4 6 2 1 6 3 0,33 3;6
reception’ and requires massive digital processing; 5 8 2 1 8 3 0,33 4;8
6 10 1 2 10 3 0,33 5;10
(b) from one specific direction 𝜃𝑖 using a narrow, high-gain 7 12 2 1 12 3 0,33 6;12
beam; 8 14 2 1 14 3 0,33 7;14
9 16 1 2 16 3 0,33 8;16
6. no simultaneous transmit and receive at a node is possible to 10 18 2 1 18 3 0,33 9;18
avoid overloading the receiver (as with traditional TDMA). 11 20 2 1 20 3 0,33 10;20
12 22 1 2 22 3 0,33 11;22
Practical beam-steering antennas conforming these characteristics 13 24 2 1 24 3 0,33 12;24
can be monolithic or be constructed of multiple modules to cover the 14 26 2 1 26 3 0,33 13;26
complete azimuth. Advantages of directional communication, and thus 15 28 1 2 28 3 0,33 14;28
16 30 2 1 30 3 0,33 15;30
of beam-steering, are the increased range at the same power level
17 32 2 1 32 3 0,33 16;32
(or equivalently, same range at less power) and increased robustness 18 34 1 2 34 3 0,33 17;34
against interference. The transmission option with low gain in the 19 36 2 1 36 3 0,33 18;36
entire azimuth loses these advantages and is therefore not consid- 20 38 2 1 38 3 0,33 19;38
ered. So only narrow, high-gain transmission in combination with
multi-direction reception is considered in the following.
MIMO is another technique that results in beam forming. However,
in the maritime domain MIMO lacks effectiveness due to the absence
of reflections and thus channel diversity. Since wireless multi-hop
networks in the maritime domain is an interesting application for the
authors, MIMO is not considered in this paper.
Networks using beam-steering antennas necessarily must use space–
time multiplexing, and thus STDMA scheduling. Like normal TDMA,
this requires each node to operate on the same frequency, and, as with
TDMA, time synchronization between all nodes is required.
With the above-described properties of beam-steering antennas, the
parameters in the expressions for capacity and delay are recalculated.

5.2. Analysis for string topology

The determining parameters for the capacity and delay in a string

Fig. 7. STDMA scheduling in a string for various frames.
topology are 𝜌, 𝑆 𝑖𝑚𝑇 𝑥, 𝑠 and 𝜆. In Fig. 6 the optimal scheduling of
transmissions between nodes with beam-steering antennas in a string
is shown. When using multi-direction reception, the scheduling can
be more efficient than in case of plain TDMA: two 𝑆 𝑖𝑚𝑇 𝑥 in three 5.3. Analysis for star topology
hops instead of two 𝑆 𝑖𝑚𝑇 𝑥 in four hops, respectively. This means that
𝜌𝑠𝑡𝑑 𝑚𝑎 = 2∕3 for beam-steering nodes. In Fig. 7 these values are used in For a 𝑁-node star topology with beam-steering, the number of time
an example STDMA scheduling for a 12-node string. Table 3, which slots differs from TDMA due to multi-direction reception. The central
is comparable to Table 2, has been constructed using the equations node can receive from all 𝑁 − 1 other nodes in the same time slot, and
defined for TDMA but now with the determining parameters 𝜌𝑠𝑡𝑑 𝑚𝑎 and the central node needs another 𝑁 − 1 time slots to transmit to these
𝜆𝑠𝑡𝑑 𝑚𝑎 for STDMA. nodes, thus 𝑠𝑠𝑡𝑎𝑟,𝑠𝑡𝑑 𝑚𝑎 = 𝑁. Again, using Eq. (4) the capacity can be
While the required number of TxOps for beam-steering nodes is the calculated.
same as for plain TDMA, i.e., 𝑀 = 2(𝑛− 1), the required number of time The delay in an 𝑁-node STDMA star is calculated in a similar way as
slots 𝑠𝑠𝑡𝑟𝑖𝑛𝑔 ,𝑠𝑡𝑑 𝑚𝑎 is less due to the higher scheduling density of STDMA. for TDMA. Given the schedule shown in Fig. 8, the number of time slots
The delay in a beam-steering string can be observed from the in a frame is simply 𝑁, and the delay in a star topology is calculated
scheduling shown in Fig. 7. With blue arrows the propagation of data with Eq. (6).
packets from node 1 to node 12 is shown, and with green arrows
propagation in the opposite direction is shown. It shows that the 5.4. Analysis for arbitrary topology
scheduling rates in both directions differ: 𝜆𝑠𝑡𝑑 𝑚𝑎 = 1 in one direction,
and 𝜆𝑠𝑡𝑑 𝑚𝑎 = 0.5 in the other direction. Applying Eq. (6) with these For STDMA, the string and star expressions can be combined in
values gives the delay for both directions as shown in the last column a similar way as for TDMA to assess the properties of an arbitrary
of Table 3. topology. Fig. 9 shows the scheduling of transmissions for the same

6
J. Laarhuis and A. Chiumento Ad Hoc Networks 169 (2025) 103750

Table 4
Relative capacity and delay on an arbitrary path for TDMA and STDMA with varying
maximum degree 𝑚𝑚𝑎𝑥 .
𝑚𝑚𝑎𝑥 TDMA STDMA
𝜌 𝜆 𝐶̃ 𝐷̃ 𝜌 𝜆 𝐶̃ 𝐷̃
2 0.5 0.5 0.25 2 0.67 1.0;0.5 0,33 1;2
3 0.5 0.33 0.17 3 0.67 0.40;0.38 0,25 3;3
4 0.5 0.25 0.13 4 0.67 0.33;0.30 0,20 3;4
5 0.5 0.20 0.10 5 0.67 0.29;0.25 0,17 4;4

Fig. 8. STDMA scheduling in a 5-node star.

results in a lower delay for beam-steering strings. On the other hand,

‘classical’ MANETs have an even distribution of forward and backward
delay over a path. In beam-steering MANETs such an even distribution
cannot be achieved.
The results for the capacity and delay of a path through a wire-
less multi-hop network with an arbitrary topology are summarized
in Table 4. An arbitrary topology characterizes as more than two
neighbours of nodes on the path, so-called non-string neighbours. To
accommodate transmissions to and from those non-string neighbours
of middle-nodes, the number of time slots in a frame is increased. The
amount of extra time slots depends on the maximum degree 𝑚𝑚𝑎𝑥 of
the middle nodes. In Table 4 the value 𝑚𝑚𝑎𝑥 = 2 corresponds to a
Fig. 9. STDMA scheduling for an arbitrary path as composition of string and star or string topology, while higher values of 𝑚𝑚𝑎𝑥 correspond to arbitrary
different additional nodes. topologies. The table shows the scheduling density 𝜌, the scheduling
rate 𝜆, the relative capacity 𝐶̃ and the relative delay 𝐷,
̃ i.e., relative
to the path length 𝑛, as a function of 𝑚𝑚𝑎𝑥 . The values for 𝑚𝑚𝑎𝑥 = 2
network as in Fig. 5. Some efficiency can be gained due to multi- correspond with the values in Tables 2 and 3. Due to the multi-direction
direction reception. The receptions from the non-string neighbours can reception, beam-steering MANETs have fewer additional time slots for
be scheduled in the same time slot as those for the string neighbours. the same maximum degree than ‘classical’ MANETs. This results in
Only 𝛥𝑠𝑠𝑡𝑑 𝑚𝑎 = 𝑚𝑚𝑎𝑥 − 2 additional time slots are needed to transmit higher capacity and lower delay for a path in beam-steering MANETs
to each of the non-string neighbours. The resulting expression for the with arbitrary topology. The table shows that the capacity decreases as
number of time slots is: the maximum degree increases. This means that the capacity of a path
𝑠𝑝𝑎𝑡ℎ,𝑠𝑡𝑑 𝑚𝑎 = 𝑠𝑠𝑡𝑟𝑖𝑛𝑔 ,𝑠𝑡𝑑 𝑚𝑎 + 𝛥𝑠𝑠𝑡𝑑 𝑚𝑎 does not depend on the total number of nodes in a network, but rather
(13) on the maximum degree of the nodes on the concerned path.
= 3 + (𝑚𝑚𝑎𝑥 − 2),
6.2. Applicability of the results
Like in the case of TDMA, the capacity of an 𝑛-hop path in an
arbitrary topology can be calculated with Eq. (4). The delay can be
obtained by first taking the additional time slots into account in a The derived approach is suitable to estimate the maximum achiev-
similar way as with TDMA using Eq. (12) with the appropriate values, able capacity and minimum achievable delay. In practice, these values
and then calculating the delay using Eq. (6). can be worse due to the following three aspects omitted in the analysis.
The first aspect is the data traffic crossing the string. While it may
6. Discussion of the results seem these flows are taken into account by the additional time slots
in Eq. (11) and Eq. (13), the applied scheduling may conflict with
6.1. Summary of the results optimal scheduling for the corresponding crossing string. It is doubtful
whether an optimal TDMA scheduling can be found for an arbitrary
The capacity and delay in a string topology are governed by two topology anyhow. Sub-optimal, heuristic scheduling will result in more
parameters: the scheduling density (𝜌) and the scheduling rate (𝜆). time slots per TDMA frame and thus leads to lower capacity values and
These parameters are entirely determined by the properties of the higher delay values than derived in this paper.
wireless network. The larger 𝜌 is, the more simultaneous transmissions
While the scheduling applies to transmissions between adjacent
occur on the string in one time slot, and thus the fewer time slots per
nodes, a routing protocol is needed to make forwarding decisions at
frame are needed, resulting in a higher capacity. For larger values of
each node. Such decisions take time, which may prevent a node from
𝜆, a packet can propagate faster through the string, thus lowering the
using the immediate available time slot to forward a packet towards
delay.
the destination. This will result in a larger delay, but will not affect the
In Table 2 and Table 3, the capacity and delay of a string are shown
capacity.
for ‘classical’ strings (TDMA) and ’beam-steering’ strings (STDMA),
respectively, as a function of the path length 𝑛. As can be seen in the A key property of MANETs is the change of topology due to move-
column 𝐶̃ in both tables, the relative capacity of the path, i.e., relative ment of constituent nodes with respect to each other. This requires a
to the link rate 𝑟, does not decrease when the path length increases, but change of TDMA frame composition every time the topology changes.
remains constant. The value of the capacity depends on the frame size Since this occurs when the network is in operation, rescheduling must
𝑠, which in turn depends on the scheduling density 𝜌. The value for 𝜌 be accomplished in real time by the nodes collectively. Where the
is higher for beam-steering strings than for ‘classical’ strings, which is scheduling of a static arbitrary topology is already difficult, dynamic
caused by multi-direction reception and a smaller contention area. This rescheduling in addition requires distribution of the topology change
means that the capacity of beam-steering strings is larger. The value for to all nodes and recalculation of the new TDMA scheduling. This poses
𝜆 is also larger for beam-steering strings than for ‘classical’ strings. This an additional complexity to (S)TDMA in MANETs.

7
J. Laarhuis and A. Chiumento Ad Hoc Networks 169 (2025) 103750

7. Conclusion Data availability

In this paper we have derived closed-form expressions for the ca- No data was used for the research described in the article.
pacity and minimum achievable delay of a path in wireless multi-hop
networks. These expressions are valid under the conditions of conflict-
References
free scheduling of time slots between neighbouring nodes, and each
node always having data to be transmitted to each of its neighbours. [1] P. Li, Capacity of wireless ad hoc networks, in: X.S. Shen, X. Lin, K. Zhang
By combining expressions for capacity and delay of elementary (Eds.), Encyclopedia of Wireless Networks, Springer International Publishing,
string and star topologies, a formulation for the maximum achievable Cham, 2020, pp. 160–164, http://dx.doi.org/10.1007/978-3-319-78262-1_64.
capacity and the minimum achievable delay have been derived for an [2] P. Gupta, P. Kumar, The capacity of wireless networks, IEEE Trans. Inform.
Theory 46 (2) (2000) 388–404.
𝑛-hop path in a MANET with arbitrary topology. The approach first
[3] R. Nelson, L. Kleinrock, Spatial TDMA: A collision-free multihop channel access
finds the conflict-free scheduling for the path as if it were a string, and protocol, IEEE Trans. Commun. 33 (1985) 934–944.
then adds time slots to this schedule to accommodate for the additional [4] S. Chen, M. Huang, S. Tang, Y. Wang, Capacity of data collection in arbitrary
transmissions required for the nodes on the path considered a star. This wireless sensor networks, IEEE Trans. Parallel Distrib. Syst. 23 (2012) 52–60.
approach has been applied to ‘classical’ MANETs, where each node has [5] A. Chou, V. Li, Slot allocation strategies for TDMA protocols in multihop packet
radio networks, in: Proc. IEEE Military Communications and Information Systems
one omni-directional antenna, and to beam-steering MANETs, in which Conference, INFOCOM, Florence, Italy, 1992, pp. 710–716.
each node can steer a narrow beam in any direction of the azimuth. [6] I. Chlamtac, A. Faragó, An optimal channel access protocol with multiple
Our results show that the capacity of a path decreases as a function reception capacity, IEEE Trans. Comput. 43 (1994) 480–484.
of the maximum degree of the nodes on the path. This is in contrast [7] G. Sharma, R.R. Mazumdar, N.B. Shroff, On the complexity of scheduling in
wireless networks, in: Proc. ACM International Conference on Mobile Computing
with many results in the literature, where the capacity of a node
and Networking,vol. 12, MOBICOM, Los Angeles, USA, 2006, pp. 227–238,
decreases as a function of the total number of the network. We thus http://dx.doi.org/10.1145/1161089.1161116.
achieve a much less conservative estimation of the capacity than what [8] B. Ji, C. Joo, N. Shroff, Throughput-optimal scheduling in multihop wire-
is usually reported in the literature. While not directly comparable, we less networks without per-flow information, IEEE/ACM Trans. Netw. 21 (2)
believe the capacity of a path is a more valuable performance metric (2013) 634–647, http://dx.doi.org/10.1109/TNET.2012.2205017, URL http://
ieeexplore.ieee.org/document/6237558/.
for users of a network than the capacity of an individual node in the
[9] J.-C. Bermond, D. Mazauric, V. Misra, P. Nain, Distributed link schedul-
network. ing in wireless networks, Discrete Math. Algorithms Appl. 12 (05) (2020)
The capacity and delay for a path in practical MANETs (for both 2050058, http://dx.doi.org/10.1142/S1793830920500585, URL https://www.
‘classical’ and beam-steering) will be less than the derived values worldscientific.com/doi/abs/10.1142/S1793830920500585.
[10] S. Chakraborty, S. Nandi, S. Chattopadhyay, Alleviating hidden and exposed
because we neglected the scheduling of and interaction with crossing
nodes in high-throughput wireless mesh networks, IEEE Trans. Wireless Commun.
paths, the routing/forwarding overhead, and the mobility and dynam- 15 (2) (2016) 928–937, http://dx.doi.org/10.1109/TWC.2015.2480398.
ics in the network. These notwithstanding, the work in this paper can
be used to determine upper bounds for arbitrary MANET networks, and
Jan Laarhuis is with Thales Nederland as of 2004. He is
the derived results will be valuable for comparison with other types of
a systems engineer at the Naval division and his primary
networks. A future research direction is how to incorporate one or more interest is distributed systems. As of 2013, Jan has a part-
of these neglected aspects into the developed approach. time appointment as a lector (professor) at the Saxion
University of Applied Science in Enschede. In this role,
CRediT authorship contribution statement he managed applied research projects. As of 2021, he is
a part-time fellow at the Pervasive Systems group of the
University Twente, where he performs research on multi-
Jan Laarhuis: Writing – original draft, Investigation, Conceptual- agent (swarm) systems. Before joining Thales, Jan has been
ization. Alessandro Chiumento: Writing – review & editing. working for fifteen years in telecommunications research
at KPN and TNO. In 1995, Jan received a PhD-degree in
Declaration of competing interest electrical engineering from the Twente University on the
topic of all-optical networking.

The authors declare that they have no known competing finan-

cial interests or personal relationships that could have appeared to Alessandro Chiumento is assistant professor at the Per-
vasive Systems group at the University of Twente since
influence the work reported in this paper.
2020, before that he was MSCA fellow at Trinity college
Dublin and he obtained his Ph.D. at KU Leuven in Bel-
Acknowledgement gium in 2015. His research interests include distributed
intelligence over networked systems, mobile computing,
The authors would like to thank Hans van den Berg from the De- autonomous multi-agent systems and socially responsible
engineering. He is particularly interested in building ex-
sign and Architecture of Communication Systems (DACS) group of the
tremely flexible, self-configurable and adaptable, wireless
University Twente for providing valuable feedback on the manuscript networks for communication systems able to adapt to strin-
of this paper. gent and conflicting objectives in terms of user satisfaction,
configurability and durability.