Professional Documents
Culture Documents
Duplexing Multiplexing and Multiple Access A Comparative Analysis For Mesh Networks
Duplexing Multiplexing and Multiple Access A Comparative Analysis For Mesh Networks
Duplexing Multiplexing and Multiple Access A Comparative Analysis For Mesh Networks
MESH NETWORKS
David W. Matolak
Ohio University
Athens, OH
2 of 7
Authorized licensed use limited to: ULAKBIM UASL - Atilim Universitesi. Downloaded on October 07,2023 at 09:38:01 UTC from IEEE Xplore. Restrictions apply.
PF = PTotal /[β(n-1)]. (1) as are comments on placement of Txij and Txji channels;
additionally, guard times are not shown.
where β is the power amplifier backoff used to avoid non-
linear distortion when a multi-carrier signal is transmitted.2 To attain an average data rate of Rb per one-way link, the
Generally 1≤β≤n-1 for n-1 carriers, with the upper limit per-slot data rate must be Rb,T=LRb, which requires the
applicable when no linearization, “pre-distortion, or peak- same total bandwidth as in the F case, W≅LRb. With PTotal
to-average power ratio (PAPR) reduction techniques are available at any node during any time slot, and recalling
used. These techniques are fairly well established, so that L=n(n-1), the transmitted bit energy is
implementations would likely have β<n-1.
Eb,T = PTotal /Rb,T= PTotal /[n(n-1)Rb]. (3)
For data rate Rb, the transmitted bit energy on any link is
… …
Eb,F = PF /Rb = PTotal /[β(n-1)Rb]. (2) Tx12 Tx21 Txij Txji Txn-1,n
Tslot t
and the received energy per bit is proportional to (2) Tf=LTslot
divided by dα. Fig. 3. Time division slot arrangement for a mesh network of L=n(n-1) links
3 of 7
Authorized licensed use limited to: ULAKBIM UASL - Atilim Universitesi. Downloaded on October 07,2023 at 09:38:01 UTC from IEEE Xplore. Restrictions apply.
f
f
f2 Tx31 Tx32 Tx23
… Tx3 f3 Tx31 Tx32
f1 Tx21 Tx12 Tx13
Tx2 f2 Tx23 Tx21 …
Rx1 Rx2 Rx3 t
Fig. 4. Illustration of time and frequency allocations for n=3 2D TF1 scheme. Tx1 f1 Tx12 Tx13
Actual assignment of frequency channels to the n-1 t
transmitting nodes during each time slot is arbitrary, but in Fig. 6. Time and frequency allocations for n=3 2D TF3 scheme.
general, at least some of the nodes must be able to both
transmit and receive on all n-1 frequency channels (during The final TF scheme is shown in Fig. 7. For this scheme
different time slots). Each frequency channel has each frequency channel is dedicated to reception by a
bandwidth approximately W/(n-1), and for average rate Rb, given user, and the slot data rates, per link transmit powers,
transmission during each slot must be at Rb,TF1=nRb. With and energy per bit are the same as that for scheme three.
available transmit power PTotal during any slot, the
transmitted bit energy is f
The second TF scheme is shown in Fig. 5, with n time Rx2 f2 Tx32 Tx12 …
slots and n-1 frequency channels, but here time slots are
assigned to transmissions by a given node, hence multi-
carrier transmission is required. The per-slot data rate is Rx1 f1 Tx21 Tx31
Rb,TF2=nRb, and the power per link is PTotal/[β(n-1)]. The
transmitted bit energy is then t
Fig. 7. Time and frequency allocations for n=3 2D TF4 scheme.
Eb,TF2 = PTotal /[β(n-1)Rb,TF2] = PTotal /[βn(n-1)Rb]. (5)
IV. COMPARISON OF DM/MA SCHEMES
f
A. Quantitative Comparison
f2 Tx13 Tx23 Tx32
… For all schemes, we can compare relative ranges by
comparing the distances at which all obtain the same
f1 Tx12 Tx21 Tx31 received bit energy, assuming the same available PTotal and
the same average data rate Rb. The received bit energy is
Tx1 Tx2 Tx3 t equal to the transmitted bit energy divided by dα. For any
of the schemes, we have
Fig. 5. Time and frequency allocations for n=3 2D TF2 scheme.
Eb,rec’d = PT/link /[Rbxdα] (7)
The third TF scheme is illustrated in Fig. 6, with n-1 time
slots and n frequency channels; each frequency channel is where the power per link in each scheme is PT/link, and the
dedicated to one user, and each node uses only single transmitted data rate is Rbx. For example, for the T and F
carrier transmission. Similar to the previous schemes, we schemes, we have
have Rb,TF3=(n-1)Rb, per link transmit power is PTotal, and
transmitted bit energy is PT
Eb ,rec 'd , F = . (8)
β (n − 1) Rb d α
Eb,TF3 = PTotal/Rb,TF3=PTotal/[(n-1)Rb] (6)
4 of 7
Authorized licensed use limited to: ULAKBIM UASL - Atilim Universitesi. Downloaded on October 07,2023 at 09:38:01 UTC from IEEE Xplore. Restrictions apply.
PT summary of the comparisons among all six schemes we
Eb ,rec 'd ,T = (9) have described is shown in Table I.
n( n − 1) Rb d α
Two rows of Table I compare the peak and average
where we’ve abbreviated PTotal by PT. Hence with equal PT,
transmit powers required of nodes in each scheme
equal average Rb and equal Eb, we have the following
(assuming a constant envelope modulation for simplicity).
range relationship between these two schemes:
One might claim that since the T scheme has a lower
1/ α
average power (because of its duty cycle), that the
dF ⎛ n ⎞ comparison is not fair. The counter argument to this is that
=⎜ ⎟ (10)
d T ⎜⎝ β ⎟⎠ all schemes require a power amplifier that can generate PT
watts for transmission. Some schemes use PT continuously,
so that dF>dT since generally β<n-1. Alternatively if we set whereas others do not. The best schemes in terms of range
the ranges of the two schemes to be equal, the data rate (or Rb) are TF3 and TF4, but TF1 is close, and does not
relationship for the two schemes is require the node to simultaneously transmit and receive.
The last row of Table I lists received SNRs, with
SNR=Prec’d/(N0B) with Prec’d equal to the per-link
Rb , F ⎛n⎞
= ⎜⎜ ⎟⎟ (11) transmitted power divided by dα, B the occupied
Rb,T ⎝β ⎠ bandwidth per transmission, and N0 the thermal noise
density.
Hence for the same received SNR, the F scheme generally
has a larger range or larger data rate than the T scheme. A
TABLE I
SUMMARY OF DUPLEXING, MULTIPLEXING/MA SCHEME CHARACTERISTICS
Duplexing/MA Scheme
Characteristic F T TF1 TF2 TF3 TF4
# time slots 1 L n n n-1 n-1
# frequency L 1 n-1 n-1 n n
channels
Node Tx duty 1 1/n (n-1)/n 1/n 1 1
cycle
Node Rx duty 1 1/n 1/n (n-1)/n 1 1
cycle
Occupied W/L W W/(n-1) W/(n-1) W/n W/n
BW/transmission
Data rate per Rb LRb nRb nRb (n-1)Rb (n-1)Rb
transmission
Transmitted power PT /[β(n-1)] PT PT PT /[β(n-1)] PT PT
per link
Energy per bit PT /[β(n-1)Rb] PT /[LRb] PT /[nRb] PT /[βLRb] PT /[(n-1)Rb] PT /[(n-1)Rb]
(transmitted)
Node peak power PT PT PT PT PT PT
Node average PT/β PT/n PT(n-1)/n PT/β PT PT
power
# carriers/Tx n-1 1 1 n-1 1 1
Simultaneous Tx yes no no no yes yes
and Rx (node)?
Relative range (for ⎛ 1 ⎞
1/ α
⎛ 1 ⎞
1/ α
⎛1⎞
1/ α
⎛ 1 ⎞
1/α
⎛ 1 ⎞
1/ α
⎛ 1 ⎞
1/ α
equal Rb , Eb) ⎜⎜ ⎟⎟ ⎜⎜ ⎟⎟ ⎜ ⎟ ⎜⎜ ⎟⎟ ⎜ ⎟ ⎜ ⎟
⎝ β ( n − 1 ) ⎠ ⎝ n( n − 1) ⎠ ⎝n⎠ ⎝ βn(n − 1) ⎠ ⎝ n −1⎠ ⎝ n −1⎠
Received SNR per PT n PT PT (n − 1) PT PT n PT n
link β d α N 0W α
d N 0W d α N 0W α
β d N 0W α
d N 0W α
d N 0W
5 of 7
Authorized licensed use limited to: ULAKBIM UASL - Atilim Universitesi. Downloaded on October 07,2023 at 09:38:01 UTC from IEEE Xplore. Restrictions apply.
Using the relationships described, our full mesh network B. Additional Remarks on Comparison
can very easily be analyzed in terms of channel capacity.
Assuming an additive white Gaussian noise (AWGN) Clearly the selection of DM/MA technique depends
channel, the one-way per link capacity in bits/second is upon more than range and data rate, even for our single
given by [19] full mesh example. One comment is that we have made
no restrictions on node complexity. The F and TF2
B schemes require each node to transmit simultaneously on
C= log 2 [1 + SNR ] (12)
J n-1 frequency channels. In practice this can make it
difficult to change the value of n unless adjacent
with B the bandwidth of the channel in Hz, J the duty frequency channels can be easily combined and/or
cycle, or equivalently, the total number of time slots, and divided. (One potential way of doing this is to employ
the SNR is the received signal-to-noise ratio. With our OFDMA, with a selectable number of subcarriers, in
assumptions on equal data rates and node capabilities, addition to RF filtering.) It is clear that changing the
the total network capacity will be L times the value in number of nodes in the T or TF schemes can be easier to
(12). For our three schemes, we have BF=W/L, BT=W, accomplish simply by the addition of more time slots,
BTF1=BTF2=W/(n-1), and BTF3=BTF4=W/n; and JF=1, with the penalty being increased latency. Practically,
JT=L, JTF1=JTF2=n, and JTF3=JTF4=n-1, so for all three nodes would likely be designed to accommodate some
schemes, B/J= W/L. maximum value of n.
From these relationships, (12), and the SNRs in Table I, In general, latency will be smallest for the F scheme, as
it is easy to deduce that CTF3=CTF4>CTF1>CF>CT>CTF2. there is no “wait time” built into this DM/MA scheme as
Fig. 8 plots C/W versus “normalized distance” d for n=3 is true for the others. The latency for the T scheme will
and α=2. For convenience in Fig. 6 we assume be largest, and that of the TF schemes in between.
PTotal/[N0W]=1, and d takes values in the range of 2 to 20. Particularly for long range applications, where guard
The ordinate scale of Fig. 6 is logarithmic simply to times in the TF and T schemes may be appreciable,
separate the curves for clarity, but the relative latency can be a problem. This is especially true for
relationships among the capacity values for all schemes some applications like voice over IP, TCP, etc., since
stay the same for any other values of n and α. For the with large guard times, to make time slots efficient in
schemes that require multi-carrier transmission, we use terms of actual data transfer versus guard times, time
the worst case value of backoff β=n-1; hence with PAPR slots must be larger, and this will increase latency. One
reduction, these schemes (F and TF2) could improve. advantage of the T scheme that is often proffered is the
The aim of the plot is not to show absolute capacity ease with which asymmetric data transfer is
values, but rather the relative values among the schemes. accomplished. Although true, in many topologies, this
means that either latency increases for those nodes not
requiring higher data rates, or that all nodes must accept
the largest duty cycle rate, as occurs for example in hub-
0
10
CF
CT
and-spoke topologies when one spoke requires say a
CTF1
50% duty cycle. Hence this advantage of the T scheme
CTF2 does not always apply.
-1
10 CTF3=CTF4
C/W (bps/Hz)
2 4 6 8 10 12 14 16 18 20
path exists from any node to any other, is on the order of
Normalized Distance d loge(n) [9]. For 20 nodes, this means any node need not
Fig. 8. Per-link relative spectral efficiency for six duplexing, multiplexing/MA be able to connect to more than ⎡log e ( 20 )⎤ = 3 other
schemes vs. normalized distance d, for a constant total transmission power
and bandwidth, n=3, and α=2 nodes at any given time. For the pure F scheme with n=3,
6 total frequency channels are required, which could be
6 of 7
Authorized licensed use limited to: ULAKBIM UASL - Atilim Universitesi. Downloaded on October 07,2023 at 09:38:01 UTC from IEEE Xplore. Restrictions apply.
feasible with development of an adaptable stop-band
filter and high-rate signal processing techniques such as
fast sampling and transmultiplexing [20]. This offers a REFERENCES
possible solution to the F scheme’s node “gender”
[1] I. F. Akyildiz, X. Wang, W. Wang, “Wireless Mesh Networks: A Survey,”
problem cited in [3]. The TF schemes require even fewer Journ. Computer Networks (Elsevier), March 2004.
frequency channels, and some are single carrier, and TF1 [2] J. D. Camp, E. W. Knightly, “The IEEE 802.11s Extended Service Set
Mesh Networking Standard,” IEEE Comm. Mag., vol. 46, no.8, pp. 120-125,
does not require simultaneous transmission and August 2008.
reception for any node. [3] J. Yang, J. Boyd, D. Laney, J. Schlenzig, “Next Generation Half-Duplex
Common Data Link,” Proc. MILCOM ’07, Orlando, FL, 29-31 October 2007.
[4] G. Aggelou, Mobile Ad Hoc Networks, McGraw-Hill, New York, NY,
Given the same power amplifier, the TF schemes may be 2005.
good alternatives to the F scheme. As with the F and T [5] M. Haenggi, D. Puccinelli, “Routing in Ad Hoc Networks: A Case for
schemes, if the number of nodes changes, the TF node Long Hops,” IEEE Comm. Mag., vol. 43, no. 10, pp. 93-101, October 2005.
[6] F. Li, Y. Wang, “Routing in Vehicular Ad Hoc Networks: A Survey,”
requires either adjustment of frequency channels or IEEE Vehicular Technology Mag., vol. 2, no. 2, pp. 12-22, June 2007.
adjustment of the number of time slots, or both. [7] P. Gupta, P. R. Kumar, “The Capacity of Wireless Networks,” IEEE Trans.
Information Theory, vol. 46, pp. 388-404, March 2000.
[8] A. Otyakmaz, R. Schoenen, S. Dreier, B. H. Walke, “Parallel Operation of
For both commercial and military applications, other Half- and Full-Duplex FDD in Future Multi-Hop Mobile Radio Networks,”
considerations also arise, such as the “tier” in which a Proc. 14th European Wireless Conf., Prague, Czech Republic, 22-25 June
2008.
node exists (e.g., “backbone” versus “edge”), the use of [9] F. Xue, P. R. Kumar, “The Number of Neighbors Needed for Connectivity
low power (e.g., handheld) terminals with more severe of Wireless Networks,” Wireless Networks (ACM Journal), vol. 10, no. 2, pp.
power constraints, and the use of directional antennas, 169-181, March 2004.
[10] C. Comaniciu, H. V. Poor, “On Energy-Efficient Cross-Layer Design:
etc. These considerations, in addition to others such as Joint Power Control and Routing for Ad Hoc Networks,” EURASIP Journ.
the study of other node topologies, use of spread Wireless Comm. & Networking, Article ID 60706, Volume 2007.
[11] K. Stamatiou, J. G. Proakis, “Assessing the Impact of Physical Layer
spectrum, and development of other, potentially more Techniques on Ad Hoc Network Performance,” Physical Communication
“global” system performance metrics [21], provide Journal (Elsevier), vol. 1, pp. 84-91, 2008.
ample area for future work. [12] G. Ferrari, S. A. Malvassori, O. K. Tonguz, “On Physical Layer-Oriented
Routing with Power Control in Ad Hoc Wireless Networks,” Proc. IET
Communications, vol. 2, no. 2, pp. 306-319, 2008.
[13] L. Schiavone, “Airborne Networking—Approaches and Challenges,”
V. CONCLUSIONS Proc. MILCOM ’04, Monterey, CA, Oct. 31- Nov. 3, 2004.
[14] D. J. Van Hook, M. O. Yeager, J. D. Laird, “Automated Topology
Control for Wideband Directional Links in Airborne Military Networks,”
In this paper, we cited examples from the large body of Proc. MILCOM ’05, Atlantic City, NJ, 17-20 October 2005.
work on mesh networks and MANETs, followed by a [15] P. Griessler, J. B. Cain, R. Hanks, “Modeling Architecture for DTDMA
Channel Access Protocol for Mobile Network Nodes using Directional
straightforward comparative analysis of several Antennas,” Proc. MILCOM ’07, Orlando, FL, 29-31 October 2007.
duplexing, multiplexing/MA schemes. When all [16] K. Karras, T. Kyritsis, M. A. Amirfeiz, S. Baiotti, “Aeronautical Mobile
schemes employ the same power amplifier, we showed Ad Hoc Networks,” Proc. 14th European Wireless Conference, Prague, Czech
Republic, 22-25 June 2008.
that the TF schemes have clear advantages in terms of [17] C. Rokitansky, M Ehammer, T. Graupl, M. Schnell, S. Brandes, S.
range and data rate over the competing schemes that use Gligorevic, C. Rihacek, M. Sajatovic, “B-AMC—A System for Future
Broadband Aeronautical Multicarrier Communications in the L Band,” Proc.
only time or frequency division (although the pure F IEEE Digital Avionics Systems Conf., Dallas, TX, 21-25 October 2007.
scheme can be competitive with effective PAPR [18] P. W. C. Chan, et. al., “The Evolution Path of 4G Networks: FDD or
reduction). We also discussed other network and node TDD?” IEEE Comm. Mag., vol. 44, no. 12, pp. 42-50, December 2006.
[19] T. M. Cover, J. A. Thomas, Elements of Information Theory, 2nd ed.,
implementation issues that indicate that despite the Wiley Publishers, Hoboken, NJ, 2006.
appeal and fairly widespread usage of time division [20] f. j. harris, Multirate Signal Processing, Prentice Hall, Upper Saddle
approaches, in many cases the use of time-frequency or River, NJ, 2004.
[21] Information Theory for Mobile Ad Hoc Networks (ITMANET) DARPA
frequency division may be preferable. IPTO program, http://www.darpa.mil/ipto/programs/itmanet/itmanet.asp
7 of 7
Authorized licensed use limited to: ULAKBIM UASL - Atilim Universitesi. Downloaded on October 07,2023 at 09:38:01 UTC from IEEE Xplore. Restrictions apply.