Advanced 5G-TCP: Transport protocol for 5G Mobile
Networks
Ivan Petrov
Sales Excellence Unit
Makedonski Telekom AD Skopje
Skopje, Macedonia
ivan.petrov@telekom.mk
Abstract—Development of the Next Generation Networks is
ongoing process. The research community is focused on
development of 5G network that will assure advanced
communication. Solutions are argued on daily basis and are
expected to be finalized around 2020. In this paper, we present an
advanced transport protocol, designed to assure efficient data
usage in the future 5G mobile networks. This transport protocol
is created to prevent congestion collapses of the backhaul traffic
and to assure ultra-high data rates up to 400Gbps.
Keywords—5G; Congestion control; Mobile networks; TCP;
Transport protocol.
I. INTRODUCTION
Research community is devoted to define future high level
network design and tries to identify all stakeholder
requirements for adapting the Internet protocol stack for 5G
mobile networks [1-5]. Virtualization of network devices will
provide platform as a service, will simplify the network
management and will assure positive cost benefit analysis. It is
natural future mobile networks to employ energy harvesting
mechanism at base station access part [6], directly related with
the radio access level in order to save energy. Optimization at
terminal application level by offloading of most energy
consuming tasks to nearby fixed servers presented as mobile
cloud computing is additional value that can be achieved and
employed at application layer in order to save terminal battery.
Compared with today’s networks, future mobile networks will
have to offer radically lower cost and energy consumption per
delivered bit. The network must offer higher data transfer rates
in static and mobile mode of operation mainly because
conventional mobile broadband applications will demand
higher capacity and higher end user data rates. Next Generation
(NGN) and Future Networks are all-IP networks, meaning that
all data, control and signaling will be carried through IP-based
communication, based on Internet technologies from network
protocol layer up to the application layer, with different
heterogeneous access technologies on the lower two layers of
the protocol stack [5]. In that manner 5G networks are
expected to be all-IP networks with higher bit rates in the
access and core parts, and delays less than 1 ms in the mobile
terrestrial access part, that requires continuing work on the
Internet Protocol stack, which is consisted of Internet Protocol
Toni Janevski
Faculty of Electrical Engineering and Information
Technologies
Ss. Cyril and Methodius University
Skopje, Macedonia
tonij@feit.ukim.edu.mk
(IP) on the network layer and Transmission Control Protocol
(TCP) and User Datagram Protocol (UDP) on the transport
layer. Our objective is to find generally applicable solutions
that improve the protocol performance over high speed
wireless and mobile access networks, while maintaining the
performance over wired links in the mobile (NGN-based) core
network. TCP is common denominator for many services;
therefore by modifying TCP [7-21] the need for applying
solutions locally can be reduced. We have designed and
evaluated version of high speed TCP protocol named as
Advanced 5G-TCP, that could be implemented in the future
networks. The paper is organized as follows. Section II
describes the basics of High Speed TCP (HSTCP). Advanced
5G-TCP is presented in section III, simulation scenario and
results are discussed in section IV. Section V concludes the
paper.
II. HSTCP BASICS
HSTCP makes TCP additive increase and multiplicative
decrease parameters to be a function of the current congestion
window when high congestion window size is achieved.
Protocol modification does not address quickly reaching of
steady state speed; it utilizes the Congestion Avoidance (CA)
phase by modifying of α and β parameters. Standard TCP
protocol is defined with the following equation:
α
w← w+ CA phase , when ACK received
w (1)
w ← w − βw in case of drop
w← w+γ Slow Start
where w is congestion window size, α is protocol increase, β is
protocol decrease and γ is slow start parameter all defined in
units of Maximum Segment Size (MSS). Default used values
of increase, decrease and slow start parameters are α=1, β=0.5
and γ=1 respectively. Steady state TCP sending rate (x) is
given with:
α (2 − β )
2 β (2)
x =
p
measured in packets per round trip time (ppt). p is the packet
drop rate and represents fraction of dropped or corrupted
packets. New protocol design process requires number of
constraints to be satisfied. The protocol should be TCP
friendly when competing with standard TCP flows and all
protocols should have fair bandwidth share. It should be
incrementally deployable, should reach high speeds reasonably
quickly when in slow start, should recover from multiple
timeouts and should perform well in case of congestion,
meaning that congestion avoidance properties should be
mproved or at least the standard one should be preserved. The
main idea of High Speed TCP is to modify the TCP response
function at very low packet drop rates in order to achieve speed
of 10Gbps. HSTCP does not change the slow start parameter γ.
It is compatible with the existing TCP’s, it uses standard
increase and decrease parameters when current packet drop rate
p is greater than some value P or when the current congestion
window is at most W packets. It uses the following switch
point values W=31 and P=0.0015 given in [9]. The target drop
rate should be chosen for a congestion window W1. The target
W1 is chosen to be 83333 packets that corresponds at 10 Gbps
sending rate when 1500 byte packets and RTT=100 ms are
assumed. The packet drop rate is calculated (P=10-k) for a
congestion window of W1 packets. Default target point values
given in [9] are W1=83000 and P1=10-7. The two points (P, W)
and (P1, W1) define the response function of high speed TCP,
when p<P. It is chosen the new protocol response function to
be linear at log-log scale defined with equation of line
log W 1 − log W (3)
log w = (log p − log P ) + log W
log P1 − log P
or written as
S
1 (4)
w = pS   W .
P
Increase and decrease parameters are set to be functions of the
current window size w when the congestion window receives
values larger than the predefined threshold of W packets.
Decrease parameter β (w ) is defined with β (W ) = 0 . 5
and β (W 1 ) = B , when w ≥ W , β (w ) is let linearly to vary
as log of w according the following equation (equation of line
on semi log scale β, w):
log (w ) − log (W )
β (w ) = ( B − 0 . 5 ) + 0 .5
log (W1 ) − log (W )
(5)
Increase parameter α (w ) can be computed using the
standard TCP equation as
w 2 2 β (w ) p (w )
α (w ) =
2 − β (w ) (6)
Choosing proper α and β values is challenging. We have
noticed that in the defined constraints when cwnd receives
value of 83000 pkt and p equals 10-7, α receives values in the
interval of 1< α ≤72 and β in the interval of 0.5> β ≥0.1. It is
obvious that improvement can be done. We choose α and β to
be functions of cwnd, and β to change in the range from 0.5 to
0.1 mainly because 0.5 represents border value used by the
traditional protocols usually for lower data rates and 0.1 is
chosen because after loss, cwnd value is decreased for 10%
when very large cwnd values are used, in our opinion there is
no use to go beyond this value. α parameter will receive values
larger than 1, and the maximal value is calculated with help of
(6) for given cwnd and p values. HSTCP authors designed the
protocol response function with help of linear interpolation
defined by two given points. Graphical presentation of w(p) is
given at Fig. 1. The red interpolation curve causes rapid cwnd
increase when packet drop rate is close to 10-3. If we choose
this kind of function it will affect TCP friendliness. Green
interpolation function will cause cwnd to increase rapidly when
p receives small values close to 10-7. From Fig. 1 we can
assume the possible β functions. The red interpolation line will
decrease cwnd more rapidly and the green line will decrease
cwnd more gradually. Decrease multiplication factor decreases
the TCP traffic so it is clear that it is wanted the traffic to be
reduced immediately when congestion has occurred, meaning
that β should be kept as large as possible; implicating that the
usage of the red interpolation line will provide improved
protocol behavior.
Fig. 1. Graphical presentation of response function possibilities
III. ADVANCED 5G-TCP
It is interesting to notice that parabola function is
generalized by rational normal curves which have the
following coordinates (x, x2, x3, …xn). Standard parabola is
obtained when n=2 and the case when n=3 is known as
twisted cubic that corresponds at TCP Cubic [11] response
function. In the theory of quadratic forms the parabola
represents the graph of the quadratic form x2. Curves y=xp for
other values of p are referred to as higher parabolas and are
treated implicitly in the form xp=kyq. In accordance with this
analysis we can find analogy with the standard TCP response
function in time domain presented as 2t/RTT, where RTT
presents packet round trip time and t is the current time or
with the cubic one written as x3 with the form of y=xp. It is
native to conclude that several TCP protocols use parabola or
higher parabola as common protocol response functions. We
have decided to use parabola function instead of linear
interpolation at log log scale in order to define w(p). If we put
series of circles in the same plane as the curve, by keeping one
of the segments constant we can construct parabola. At Fig.2
is presented construction of parabola with help of series of
tangent circles passing through the point S. The segment SA is
kept constant during the construction. The fixed horizontal
line at A constructs geometric means between SA (constant)
and the series of segments AT, AU, AV. These segments are
plotted against the series of geometric means AX, AY, AZ to
give the points A, B, C and D all of which lie along the
parabola. In this picture the vertical distances of the points (B,
C and D) from A are proportional to the square on their
horizontal distances. Knowing how to construct parabola with
help of this method we came to idea to present the TCP
response function in the same manner. We have analyzed
HSTCP [8] in details and have decided instead to use the
equation of line at log log scale for known two points defined
with congestion window value (cwnd, w) and packet drop rate
(p) to construct function of parabola uniquely defined with
known three points.
We have named this protocol Advanced 5G-TCP and have
decided to keep the predefined switch point used by HSTCP
defined with w=38 and p=10-3. In order to construct parabola
we have used the following points w1=8300 pkt and p1=10-4
(corresponds at speed of 1Gbps), w2=830000 pkt and
p2=10-7 (corresponds at speed of 100Gbps), w3=3330000 and
p3=10-12 (corresponds at speed of 400Gbps). These points are
chosen because 1Gbps is expected to be common data transfer Fig. 2. Design of parabola with help of circles
speed of the 5G mobile networks, 100Gbps is expected to be
backhaul link speed and 400Gbps is the theoretical speed of
optical link defined with IEEE P802.3bs standard. If we have
this standardization in mind then we can decade to use the
future 5G network packet loss probability values and data rates
to define our switch points. Recall that the suggested bit error
probabilities are real if we use 1500 byte packets.
We have tried with these predefined points to calculate the
constant distance SA and to use it to plot the parabola function
with help of one point tangent circles. Knowing that the
process can be translated in algorithm that can calculate the
next cwnd value was promising until we found that for this set
of points the obtained SA values differed mainly because the Fig. 3. Graphical presentation of f1, f2,f3,f4,f5
chosen set of points defines the function f1 presented at Fig.3
which does not corresponds with the one presented at Fig.2
After we have obtained the result we decided to use the
standard parabola equation to define f1 graphically and
analytical with help of (7) or (8)
y = ax 2 + bx + c (7)
log (w ) = a (log ( p )) + b log ( p ) + c (8)
2

where y is log(cwnd) and x is written as log(p). For known

three points (defined with w, p; w2,p2 (100Gbps) and w3,p3
(400Gbps)) we can unique define function of parabola and if
we solve the system of three equations we will define the
values of a, b and c. Following values were calculated Fig. 4. Graphical presentation of eqiation (13)
a=-0.108, b=-2.17 and c=-3.96 which define the function f1
presented at Fig.3 We have calculated the values of a=-0.41,
b=-5.21 and c=-10.47, in the case when the points of interest
are (w,p), (w1,p1) and (w2,p2) that construct f2 which cuts f1
at point B.Obtained functions f1 and f2 are not adequate to be
used mainly because for very small values of p (p<10-12 case
when f1 is used and p<10-7 case when f2 is used) cwnd value
starts to drop which is not desired. We have decided to solve
the system of equations once more but this time for points that
correspond at speed of 5Mbps when p=10-3, 10Mbps when
p=10-4 and 100Mbps and p=10-5. Following parameter values Fig. 5. Graphical presentation of β(w) at semi log system
were calculated a=0.325, b=1.92, c=4.42 which define the
third function f3 presented at Fig.3 which cuts f1 in point A We have plotted two additional functions with help of
when cwnd corresponds at rate of 53Gbps (w4) and p4=10-6.58. equation of line, case when cwnd corresponds at rate of
We found that f3 can be constructed and plotted with help of 400Gbps, f4 and f5 that cut f1 at different p values, the first
the method described above. one in point C (p5=10-8) and the other one in point D (p5=10-
12
). From Fig.3 we conclude that improvements of the
response function choice can be done in accordance with the
given directions. Final Advanced 5G-TCP response function is
defined with f3 when w<cwnd<w4 and p4<p<p, when
w4<cwnd<w5 and p5<p<p4 the protocol will use f1 as a
response function. For values larger than w5 or when p is
smaller or equal at p5 the protocol will use the function f4.
The equation of line can be easily transformed from log
domain but this does not stand for the equation of parabola.
Important for this protocol is that it has defined response
function for low and high speed data rates and it is TCP
friendly. The standard parabola is defined with
x 2
= ky
or
x2 = y (10)
In log log scale we will have
(log ( p )) 2 = log (w ) (11)
log (w )
log ( p ) = ≡ log p (w ) (12)
log ( p )
w = p log ( p ) (13)
In the case when equation (9) is used we will have
1
log( p )
w= p k
(14)
If we plot equation (13) we will obtain the representation at
Fig.4 from where it is clear that for very small values of p, w
enlarges its value and asymptotically nears the y axis. The
function can be plotted for known a, b and c parameters.
Equation (6) is used to calculate α, while β can be described
with parabola function passing through 0.1 when
w=3330000, p=10-12; 0.5 when w=38 and p=10-3; 0.4 when
w=833pkt (100 Mbps).
For cwnd values larger than 3330000 pkt β will be
calculated according the equation of line defined by w and w3.
log(w) − log(38) (15)
β (w) = (0.1 − 0.5) + 0.5
log(3330000) − log(38)
Following values were obtained for a=-0.003, b=-0.05 and
c=0.59. Graphical presentation of β(w) at semi log system is
given at Fig.5. From Fig.5 we can conclude that we will obtain
similar results if we use equation (15) to calculate β(w). It is
important to note that variance during β(w) calculation can be
noticed for lower cwnd values so we advise parabola function
to be used in that case.
(9)
Fig. 6. Simulation scenario. We define wireless medium between n0 and the
micro base station n1. n1 is wirelessly connected with the macro base
station n2 and optical link n2-n3 is defined between the macro base
station n2 and the network gateway n3
IV. SIMULATION SCENARIO
Analysis is conducted in order to justify the design of
Advanced 5G-TCP. We use network simulator (ns2) upgraded
with Advanced 5G-TCP module. Fig.6 presents
communication of one mobile terminal with the base station
and backhaul high speed link connected with adequate
network gateway.
Cwnd variation
4500
HSTCP Cwnd
4000
5G-TCP Cwnd
Reno Cwnd
3500
3000
Cwnd (pkt)
2500
2000
1500
1000
500
0
50.002 200.01 400.01 600.02 800.03 1000 1200 1400.1 1600.1 1800.1 1990.1
Time (sec)
Fig. 7. Congestion window (cwnd) change for Gbps mobile connections with
adequate buffer sizes of 8300, 83000 and 166000 packets.
Parameter of interest is the congestion window size
variation. Cwnd is directly related with the achieved
throughput. Packet size is set at 1500 bytes. Simulation run is
2000 sec. Buffers are Drop Tail. Buffer size is set at 100% of
the product of bottleneck capacity by the largest RTT divided
by packet size. The result is presented at Fig.7. At Fig.7 we
present w as a function of time. It can be noticed that Reno
protocol has smallest window growth which directly impacts
the throughput that can be achieved when this protocol is used.
HSTCP cwnd change compared with Reno is more than
doubled over the time. Advanced 5G-TCP presents improved
results when compared with HSTCP for sessions longer than
50 seconds. As we have described in the previous section
Advanced 5G-TCP employs fast growing function that makes
it superior over long TCP connections that are expected to be
common in 5G mobile scenario.
 V. CONCLUSION
In this paper we have presented novel high speed TCP
algorithm with help of detail analysis of HSTCP. Increased
friendliness and higher data rates are achieved when Advanced
5G-TCP is employed in the predefined simulation
environment. This protocol is designed to be used in the future
5G data networks regardless of the engaged MAC Layer. With
help of several response functions that are used for variety of
speeds data transfer up to 400Gbps is assured. Today it is
essential the high speed transport protocol development to
follow the 5G network capacity in order the end user to be able
to benefit it. Novel algorithms should be created in order data
rates in the range from 1Gbps to 50Gbps easily to be assured.
Advanced 5G-TCP represents novel high speed protocol that
uses parabola in order to define the protocol response function
and the calculation of β(w) parameter. If the network capacity
allows this protocol could easily reach rates gather than
400Gbps. Further design improvements can be evaluated.
[17] D. X. Wei, C. Jin, S. H. Low, and S. Hegde, “FAST TCP: motivation,
architecture, algorithms, performance,” IEEE/ACM Trans. Netw., vol.
14, no. 6, pp. 1246–1259, 2006.
[18] J. Sing and B. Soh, “TCP New Vegas: Improving the Performance of
TCP Vegas Over High Latency Links,” in Proc. 4th IEEE International
Symposium on Network Computing and Applications, 2005, pp. 73–80.
[19] K. Yamada, R. Wang, M.Y. Sanadidi, and M. Gerla” TCP Westwood
with Agile Probing:Dealing with Dynamic, Large, Leaky Pipes” IEEE
Communication Society, 2004.
[20] D. Kliazovich, F. Granelli, and D. Miorandi, “Logarithmic window
increase for TCP Westwood+ for improvement in high speed, long
distance networks,” Computer Networks, vol. 52, no. 12, pp. 2395–
2410, August 2008.
[21] K. Kaneko, T. Fujikawa, Z. Su, and J. Katto, “TCP-Fusion: a hybrid
congestion control algorithm for high-speed networks,” in Proc.
PFLDnet, ISI, Marina Del Rey, California, February 2007.

REFERENCES
[1] M.Duke, R.Braden,W.Eddy, E.Blanton, A.Zimmermann, “A Roadmap
for Transmission Control Protocol (TCP)”. Internet Engineering Task
Force (IETF), RFC7414, February 2015.
[2] T. Janevski, “Internet Techologies for Fixed and Mobile Networks”,
Artech House, USA, November 2015.
[3] J. Rodriguez, “Fundamentals of 5G Mobile Networks”, John Wiley and
Sons, UK, 2015.
[4] X. Ge, H. Cheng, M. Guizani, T. Han, “5G wireless backhaul networks:
challenges and research advances”, Network IEEE, Nov.-Dec. 2014
[5] T. Janevski, “NGN Architectures, Protocols, and Services”, John Wiley
and Sons, UK, 2014.
[6] D. Gunduz, K. Stamatiou, N. Michelusi, and M. Zorzi, ”Designing
intelligent energy harvesting communication systems,” IEEE Commun.
Magazine, vol. 52, no. 1, pp. 210 - 216, Jan. 2014
[7] T. Kelly, “Scalable TCP: improving performance in highspeed wide area
networks,” Computer Communications Review, vol. 32, no. 2, April
2003.
[8] S. Floyd, “HighSpeed TCP for large congestion windows,” RFC 3649,
2003.
[9] L. Xu, K. Harfoush, and I. Rhee, “Binary increase congestion control for
fast, long distance networks,” in Proc. IEEE INFOCOM, vol. 4, March
2004, pp. 2514–2524.
[10] D.Leith, “H-TCP: TCP congestion control for high bandwidth-delay
product paths,” IETF Internet Draft, http://tools.ietf.org/html/draftleith-
tcp-htcp-06, 2008.
[11] I. Rhee and L. Xu, “CUBIC: a new TCP-friendly high-speed TCP
variant,” SIGOPS Operating Systems Review, vol. 42, no. 5, pp. 64–74,
July 2008.
[12] G. Marfia, C. Palazzi, G. Pau, M. Gerla, M. Sanadidi, and M. Roccetti,
“TCP Libra: Exploring RTT-Fairness for TCP,” UCLA Computer
Science Department, Tech. Rep. UCLA-CSD TR-050037, 2005.
[13] C. Caini and R. Firrincieli, “TCP Hybla: a TCP enhancement for
heterogeneous networks,” International J. Satellite Communications and
Networking, vol. 22, pp. 547–566, 2004.
[14] A. Baiocchi, A. P. Castellani, and F. Vacirca, “YeAH-TCP: yet another
highspeed TCP,” in Proc. PFLDnet, ISI, Marina Del Rey (Los Angeles),
California, February 2007.
[15] R. King, R. Baraniuk, and R. Riedi, “TCP-Africa: an adaptive and fair
rapid increase rule for scalable TCP,” in Proc. IEEE INFOCOM, vol. 3,
March 2005, pp. 1838–1848.
[16] K. Tan, J. Song, Q. Zhang, and M. Sridharan, “A compound TCP
approach for high-speed and long distance networks,” July 2005.