A Statistical Approach to Adaptive Playout Scheduling in Voice Over Internet Protocol Communication

International Journal of Electrical and Computer Engineering (IJECE)
Vol. 8, No. 5, October 2018, pp. 2926~2933
ISSN: 2088-8708, DOI: 10.11591/ijece.v8i5.pp2926-2933  2926
Journal homepage: http://iaescore.com/journals/index.php/IJECE
A Statistical Approach to Adaptive Playout Scheduling in Voice
Over Internet Protocol Communication
Priya Chandran1
, Chelpa Lingam2
1
BVIMIT, University of Mumbai, R&D Centre Bharatiar University Coimbatore, India
2
Department of Computer Engineering, Pillai HOC College of Engineering and Technology, University of Mumbai, India
Article Info ABSTRACT
Article history:
Received Nov 30, 2017
Revised Jan 13, 2018
Accepted Sep 5, 2018
Factors like network delay, latency and bandwidth significantly affect the
quality of communication using Voice over Internet Protocol. The use of
jitter buffer at the receiving end compensates the effect of varying network
delay up to some extent. But the extra buffer delay given for each packet
plays a major role in playing late packets and thereby improving voice
quality. As the buffer delay increases packet loss rate decreases, which in
general is a very good sign. However, an increase of buffer delay beyond a
certain limit affects the interactive quality of voice communication. In this
paper, we propose a statistical framework for adaptive playout scheduling of
voice packets based on network statistics, packet loss rate and availability of
packets in the buffer. Experimental results show that the proposed model
allocates optimal buffer delay with the lowest packet loss rate when
compared with other algorithms.
Keyword:
Jitter buffer management
Jitter
Playout scheduling
PLC
VoIP
Copyright © 2018 Institute of Advanced Engineering and Science.
All rights reserved.
Corresponding Author:
Priya Chandran,
BVIMIT ,University of Mumbai,
R&D Centre Bharatiar University Coimbatore, India.
Email: priyaci2005@gmail.com
1. INTRODUCTION
Voice over Internet Protocol (VoIP) is a technology used for the voice transmission over internet
protocol. The wide-spread usage of internet and cheap communication cost made VoIP a very popular
technology. In VoIP, voice conversations are digitized and then packetized for transmission. These packets
are transmitted every 20 to 40 ms [1] and expect that these packets reach the receiver side in the same
interval. If the packets are arrived at the receiver side without any delay variations, they can be played
directly. But due to network impairments like delay and latency, the packets are lost or may not arrive in the
same order in which the sender sent. This causes the end users to experience a broken, buzzing and delayed
speech. Extensive research is going on this area to increase the quality of communication [2-4]. Many
researchers proposed receiver side Packet Loss Concealment (PLC) methods [5].
The difference in packets inter-arrival time is known as jitter and to mitigate the effect of jitter, a
jitter buffer is used at the receiver side. This jitter buffer holds received packets till its estimated playout time,
to accommodate late packets. Thus the playout time of received packets are delayed to reduce packet loss
rate. Increasing buffer delay extends playout time of the packets received and hence communication
interactivity is reduced. At the same time, this reduces late packet loss rate. But decrease in buffer delay
increases late packet loss rate. Many playout scheduling algorithms have been proposed to solve this trade-
off. Playout scheduling schemes are classified as static (fixed) and adaptive. Static playout scheduling is the
simplest method, where the playout time of all the packets in a session is fixed irrespective of the varying
network delay [6]. Since the network conditions are volatile, this method is not very effective in reducing
packet loss rate. Keeping a fixed playout scheduling time for all the packets in face of varying network
conditions is not a good option.

Int J Elec& Comp Eng ISSN: 2088-8708 
A Statistical Approach to Adaptive Playout Scheduling in Voice Over… (Priya Chandran)
2927
In adaptive playout scheduling algorithms, the playout time of the packets is adaptively adjusted
according to the network delay. i.e., Packets of the same talk spurt experience different playout time and the
sequential playout of packets is implemented by PLC mechanisms like extending or compressing the silence
periods. PLC approaches like signal reconstruction and timescale modification methods are also widely used
for the sequential playout of packets at receiver side. The efficiency of adaptive playout scheduling
mechanisms depends on the accurate estimation of the buffer delay needed by the arrived voice packets so
that late packets can also be played. In this paper, we propose a framework for adaptive playout scheduling of
voice packets by estimating buffer delay of the packets based on the network statistics, packet loss rate and
availability of packets in buffer.
Several methods have been developed for estimating the adaptive playout time of packets. Ramjee
et al. [7] proposed four algorithms for adaptive playout delay estimation of voice packets. In this method, the
playout time of first packet of the talkspurt is estimated using an autoregressive estimate and an offset is
added to calculate the playout time of successive packets in the same talkspurt. But the delay adaptation is
applied only to the first packet of the talk spurt. Also the buffer size is exponentially increased to store late
packets. Pinto et al. [8] proposed an adaptive gapbased algorithm that can be tuned for both end-to-end delay
and packet loss to satisfy a user-desired tolerance.
The method proposed by Liu et al. [9] adaptively adjusts jitter buffer according to an indication of a
sequence number of the delaying packet. Approaches based on order statistics based estimation [10], network
variations [11-12] and quality-driven playout buffer optimizations [13] are also proposed.
B. H. Kim et al. [12] used a static buffer size of 200ms to hold packets. Event counting algorithm proposed
in [14] relies on the measure of relative times in handling the object-related events. Normal approximation is
used to identify the delay of network in [15]. Modified enhanced normalized least mean squares algorithm
(ENLMS) is also used to identify the spike state of the network [16].
Different methods have been proposed by researchers to solve the issues with adaptive playout
scheduling in VoIP communication. S.B.Moon et al., [17] computed upper and lower bounds on the
minimum average playout delay for a given packet loss by mainitaining delay percentile information. M. K.
Ranganathan and L. Kilmartin S [18] proposed a novel fuzzy trend analyzer system to estimate intra-talk
spurt playout delay adaptation. Sreenan and Cormac [19] suggested histogram based approaches for
synchronizing networked multimedia streams.
2. PROPOSED METHOD
In our proposed method, we perform the following operations to implement adaptive playout
scheduling.
a. Computation of network statistics
b. Estimation of playout time based on optimal buffer delay
In the following section, a brief explanation of adaptive playout scheduling is given.
2.1. Adaptive Playout Scheduling
In adaptive playout scheduling algorithms, the playout time of the packets are adaptively adjusted
during silence periods and within talkspurts. This is mainly achieved by scaling of voice packets within
talkspurts and keeping the synchronous playout of packets. Each packet in the talkspurt experiences different
playout time. Buffer delay estimation is a major component in calculating the playout time of packets. The
tradeoff between buffer delay and late packet loss rate can be solved using an efficient buffer delay
estimation method. We assume that both sender and receiver maintain clock synchronization. As
packetization delay and codec delay are codec dependent, they both are not taken into account while
computing the playout time of the packets. Playout time of packet i, tpi is the time packet is played at the
receiver and network delay, dni, is the time gap between sending and receiving time of a packet.
𝑡𝑝𝑖 = 𝑡𝑟𝑖 + 𝑑𝑏𝑖 (1)
𝑑𝑛𝑖 = 𝑡𝑟𝑖 − 𝑡𝑠𝑖 (2)
where, 𝑡𝑟𝑖, 𝑑𝑏𝑖 and 𝑡𝑠𝑖 are receiving time, buffer delay and sending time respectively. Buffering time is
estimated for each packet in a talk spurt to calculate the playout time. When the buffering time increases,
playout time of the packet also increases. Even though the packet loss rate decreases due to the increase in
buffer delay, it reduces the voice quality and listeners experience a broken, buzzing and delayed speech.
Total delay experienced by a packet, 𝑑𝑡𝑖 , is the difference between the time it is sent from sender and the

 ISSN: 2088-8708
Int J Elec& Comp Eng, Vol. 8, No. 5, October 2018 : 2926 - 2933
2928
time it is played on the receiver side. It is the sum of network delay and buffering delay and is given in
Equation (3).
𝑑𝑡𝑖 = 𝑑𝑛𝑖 + 𝑑𝑏𝑖 (3)
The jitter of ith
packet, 𝑗𝑖, for a talk spurt is calculated as,
𝑗𝑖 = 𝑑𝑛𝑖 − 𝑑𝑛𝑖−1 = (𝑡𝑟𝑖 − 𝑡𝑠 𝑖 ) − (𝑡𝑟𝑖−1 − 𝑡𝑠𝑖−1) (4)
2.2. Computation of Network Statistics and Optimal Buffer Delay
The main idea behind our new approach is to compute the playout time of a packet using buffering
delay based on the network statistics of already arrived packets. We use a vector to store the details of
already arrived packets. In order to track the latest network conditions, old packet information are removed
periodically from vector and updated with latest packet information. It follows a window based approach so
that the estimations always adapt to the varying network conditions. In addition to the past network statistics,
packets waiting in buffer and late packet loss rate are also considered as factors for buffer delay computation.
Normal and spike modes of network states for each packet are identified based on the network
delay, 𝑑𝑛𝑖, of the packets and the threshold value, 𝑡ℎ 𝑠𝑝𝑖𝑘𝑒 [12]. If a packet is received with a delay greater
than 𝑡ℎ 𝑠𝑝𝑖𝑘𝑒, current mode is identified as spike. Otherwise, the mode is taken as a normal. The network
delay is estimated using equation (5):
𝑑𝑛𝑖 = ∝ 𝑛 ∗ 𝑑𝑛𝑖−1 + (1 − ∝ 𝑛 ) ∗ 𝑑𝑛𝑖 (5)
where ∝ 𝑛 is a weighing factor which depends on delay statistics [7]. Using the above estimated network
delay, playout time of packet i is computed as given in equation (6).
𝑡𝑝𝑖 = tsi + 𝑑𝑛𝑖 + db𝑖 (6)
The accurate estimation of buffer delay for playout time calculation helps in allocating late packets in the
buffer for playing, and thereby reducing late packet loss rate. For this, we have made the value of db𝑖
proportional to the fluctuating network delay, late loss rate and availability of packets in buffer.
Buffer delay is estimated using equation (7).
db𝑖 = 𝜔 ∗ Ji ∗ Llate (7)
𝜔 is the delay factor which depends on availability of preceding and succeeding packets. The network jitter J
of i th packet is estimated as,
Ji = mi + φi
∗ vari (8)
where φ is the weighting factor of the inter-arrival jitter variance, m and var are the average and variance of
inter-arrival jitter, respectively. For every packet that arrive at the receiver, the algorithm checks the current
and previous mode, and the value of m and var is calculated differently in each mode using separate
equations as given in [12]. This way, quickly varying network conditions can be adapted more accurately for
the estimation of playout time. The value of φ is proportional to network characteristics and is calculated as:
φi
=
mi−1− σi−1
vari−1
(9)
where σ is the standard deviation. The value of m , var and σ are calculated for all the packets stored in
window in order to predict the current network state. The size of window is specified in section 3. In addition
to the packet loss rate, our method also considers order of already received packets while predicting buffer
delay of ith
packet. The calculated jitter can be negative if the packets are received out of order. We use a
delay factor 𝜔 to adjust buffer delay of expected packet and its value varies based on the above specified
scenarios. The value of 𝜔 is 𝑙𝑜𝑤, 𝑙𝑜𝑤𝑒𝑟, ℎ𝑖𝑔ℎ or ℎ𝑖𝑔ℎ𝑒𝑟 according to the availability of preceding and
succeeding packets. For example, if we check the availability of only succeeding and preceding packets
received, we assign a value to 𝜔 based on the following four different cases.
Both 𝑝𝑖−1 and 𝑝𝑖+1 are received, 0 < 𝑙𝑜𝑤𝑒𝑟 < 1

2929
𝑝𝑖−1 is received and 𝑝𝑖+1 not received, 0 < 𝑙𝑜𝑤 < 1
𝑝𝑖−1 not received, but 𝑝𝑖+1 is received, 1 < ℎ𝑖𝑔ℎ < 2
𝑝𝑖−1 and, but 𝑝𝑖+1 not received, 1 < ℎ𝑖𝑔ℎ𝑒𝑟 < 2
This way the value assigned to 𝜔 is considered as one parameter for controlling the buffer delay of packets.
For example, if both the succeeding and preceding packets are already arrived, then a small value is assigned
to 𝜔. The proposed algorithm for estimating playout time of packets is given as:
Algorithm: Estimation of Playout time
1. Receive packets
2. for (each packet in the talkspurt)
3. Compute network delay
4. Estimate the jitter and delay factor
5. Estimate buffer delay
6. Calculate playout time
7. end for
Packets are discarded based on playout time of late packets and is given as:
store 𝑝 𝑘 in buffer If ( 𝑡𝑟𝑘 < 𝑡𝑝 𝑘 )
discard 𝑝 𝑘 Otherwise
3. RESULTS AND ANALYSIS
We have evaluated our proposed algorithm for playout time estimation using an experimental setup
[20] as shown in Figure 1. The four modules used for this experimental setup are: Session Initiation Protocol
(SIP) application, network traffic emulator, VoIP server and a packet analyzer. We have tested VoIP
communication between two endpoints, endpoint1 and endpoint2. A VoIP server installed on a computer is
used to test the call flows between two end-points. One endpoint is a mobile device which is registered with
the server using a VoIP application. PJSIP application running on a computer is used as another endpoint.
Program for Network statistics based playout time estimation is written in C language. Network emulator is
used to set different network parameters like jitter, delay, packets reordering and packets loss in RTP packets
flow. The two endpoints act as SIP clients and call flows between these two SIP clients were analyzed using
packet analyzer.
Figure 1. Experimental setup
The Real Time Protocol (RTP) packets sent from endpoint1 is redirected through VoIP server (SIP
proxy server) and then sent to endpoint2. Recorded speech sample database, provided by the International
Telecommunication Union [21] are taken for our experimental study. The performance of our method is
evaluated using the metric, average buffering delay ( db 𝑎𝑣𝑔) and late packet loss rate ( L 𝑙𝑎𝑡𝑒).
db avg =
1
P
∑ dbi
n
i=1 (17)
where 𝑃 is the set of packets played.
L late =
|R−P|
R
(18)
where 𝑅 the set of packets received.

 ISSN: 2088-8708
2930
Also to evaluate communication quality, objective voice quality test is carried out using
E-model [22]. The quality of transmission rating in E-Model varies between 0 and 100 and is specified by
Transmission Rating factor (R factor). 0 represents bad quality and 100 represents extremely good quality of
communication. Then, R factor is mapped to MOS (Mean Opinion Score) scores for finding conversational
quality.
We have used network emulator to create traces with varying jitter. One way-delay of the
communication is measured for each packet having 160 bytes of payload size and the G.711 coded voice
packets are generated at the sender in 30 ms time interval. Performance of our proposed method is compared
with two different adaptive playout scheduling schemes. Method 1 uses an autoregressive estimate to predict
the network delay and jitter and is given as algorithm 4 in [7], method2 uses self adaptive jitter buffer
adjustment mechanism discussed in [9] and method 3 is based on linear prediction with a static buffer
size [12]. We denote auto_reg, self_adapt, linear_pred and PM for method1, method2, method3 and proposed
method respectively.
Playout delay estimation algorithm is executed for every packet in the talkspurt. Since the
packetization interval of audio packets is taken as 30ms [1], the algorithm should be efficient to execute
maximum 34 times per second. Also the calculation complexity increases as window size increases while
calculating network statistics. At the same time, as the window size decreases, the adaptation is more
responsive to the volatile network behavior [7] [14]. So it is important to select the window size which adapts
and estimates accurate network characteristics. After experimenting with different window sizes, we kept a
window of size 100 packets which resulted in optimal performance. For delay factor calculation, we have
checked availability of three succeeding and two preceding packets. Six speech samples were collected for
analysis, and Table 1 depicts network trace statistics along with the experimental results. Among the six
traces collected, trace1 has small jitters; traces 2, 3 and 4 are having medium jitters, while trace 4 and 6 is
having large jitters. Standard Deviation (STD) of network delay of the voice data varies for traces. The
maximum jitter experienced in trace1, trace2, trace3 trace4, trace5 and trace6 are 51, 125, 160, 180, 240 and
325 respectively. For the small jitter case (trace1), average buffer delay is nearer to the ideal case (30ms) and
as the jitter increases, buffer delay also increases. Performances of six algorithms for six different traces are
depicted in Figure 2. For the trace with larger jitter value, 325ms, proposed method gives a high MOS score
value compared with other three algorithms.
Table 1. Experimental Results
Trace
STD of network
delay (ms)
Maximum
Jitter (ms)
Methods
used
db 𝑎𝑣𝑔 L late (%)
MOS
1 9.45 51
auto_reg 42.41 5.46 3.41
self_adapt 38.53 2.67 3.51
linear_pred 38.73 2.13 3.78
PM 35.15 1.32 4.01
2 23.26 125
auto_reg 55.13 7.23 2.91
self_adapt 49.38 4.25 3.43
linear_pred 48.21 3.97 3.52
PM 44.37 2.53 3.72
3 24.11 160
auto_reg 56.28 8.62 2.99
self_adapt 48.46 4.67 3.76
linear_pred 49.84 5.14 3.47
PM 44.72 3.63 3.61
4 25.67 180
auto_reg 56.48 9.01 2.99
self_adapt 50.76 8.58 3.76
linear_pred 50.24 8.21 3.47
PM 44.97 7.03 3.61
5 27.13 240
auto_reg 61.24 9.27 2.42
self_adapt 55.72 8.24 2.63
linear_pred 54.57 10.67 2.75
PM 51.36 7.72 3.02
6 29.81 325
auto_reg 79.63 12.82 2.17
self_adapt 72.52 10.15 2.64
linear_pred 72.41 11.21 2.62
PM 64.56 8.33 2.95

2931
The performance of four adaptive playout scheduling algorithms for all the traces is depicted in
Figure 2. In all the cases proposed method results in lowest late loss rate and buffering delay. As the jitter
increases, average buffering delay is increased to reduce late packet loss rate.
Figure 2. Performance of four adaptive playout scheduling algorithms for six different traces
The comparison of late packet loss and average buffering delay estimation of algorithms for traces is
depicted in Figure 3(a) and Figure 3(b) respectively. In all the cases proposed method results in lowest late
loss, buffer loss and buffering delay.

 ISSN: 2088-8708
2932
(a) (b)
Figure 3. (a) Late packet loss rate (b) Average buffering delay of six traces
Figure 4 shows MOS score of all the traces with small, medium and large jitter cases. MOS score
obtained from the experiments indicates that for all the traces speech quality is significantly improved by
proposed algorithm. Experimental results show that proposed method estimates playout time of packets to
match the jitter variations while keeping buffering delay minimum for maintaining interactive voice quality.
Figure 4. MOS score of different algorithms
4. CONCLUSION
Widespread use of internet replaces traditional telecommunication system by the VoIP services.
Rendering high quality service to the users is a crucial challenge faced. Network delay is one major
impairment factor which affects the quality of communication. Network delay causes the late arrival or loss
of packets and there by exacerbating the voice quality. In this paper, we propose a model based on network
statistics, packet loss rate and packet availability in the buffer to estimate playout time of each packet. We
have compared our algorithm with existing three adaptive playout scheduling algorithms and the results
indicate that our method solves the trade-off between buffer delay and packet loss in a better way than the
existing adaptive playout scheduling methods with minimum packet loss rate and buffering delay. Also the
communication quality is evaluated using an objective voice quality test, E-model and our algorithm achieves
a higher MOS score, for all the traces, than the other three algorithms.
REFERENCES
[1] ITU-T Recommendation G.114. One way transmission time. 2000.
[2] Kolhar, Manjur, Mosleh M. Abualhaj, and Faiza Rizwan. QoS Design Consideration for Enterprise and Provider's
Network at Ingress and Egress Router for VoIP protocols. International Journal of Electrical and Computer
Engineering (IJECE). 2016: 6.1: 235.
[3] Audah, L., Sun, Z. Cruickshank, H. QoS based Admission Control using Multipath Scheduler for IP over Satellite
Networks. International Journal of Electrical and Computer Engineering (IJECE). 2017; 7.6: 2958-2969.
[4] Wheeb, Ali Hussein. Performance Evaluation of UDP, DCCP, SCTP and TFRC for Different Traffic Flow in
Wired Networks. International Journal of Electrical and Computer Engineering (IJECE). 2017; 7.6: 3552-3557.

2933
[5] Maheswari, K., and Punithavalli, M., Receiver based packet loss replacement technique for high quality VoIP
streams, In Nature & Biologically Inspired Computing, NaBIC 2009. World Congress on, IEEE, pp. 1669-1672.
[6] Alvarez-Cuevas, Felipe, et al. Voice synchronization in packet switching networks. IEEE Network. 1993; 7.5: 20-
25.
[7] R. Ramjee, J. Kurose, D. Towsley and H. Schulzrinne.Adaptive playout mechanisms for packetized audio
applications in wide-area networks. Proc. IEEE INFOCOM, Toronto, Canada. 1994.
[8] Pinto, Jesus, and Kenneth J. Christensen. An algorithm for playout of packet voice based on adaptive adjustment of
talkspurt silence periods. Local Computer Networks, LCN'99. Conference on. IEEE. 1999.
[9] E Liu, G Shen, S Jin, L Gui. Self-adaptive jitter buffer adjustment method for packet-switched network. U.S.10 July
2012; Patent No. 8, 218, 579.
[10] Y.J.Liang, N. Farber, and B. Girod. Adaptive playout scheduling and loss concealment for voice communication
over IP networks. IEEE Transactions on Multimedia. 2001; Vol. 5(4): pp. 532-543.
[11] L Repele, R Muradore, D Quaglia. Improving performance of networked control systems by using adaptive
buffering. IEEE Transactions on Industrial Electronics. 2014; 61.9: 4847-4856.
[12] Byeong Hoon Kim, Hyoung-Gook Kim, Jichai Jeong and Jin Young Kim. VoIP receiver-based adaptive playout
scheduling and packet loss concealment technique. IEEE Transactions on consumer Electronics. 2013; 59.1: 250-
258.
[13] Sun, Lingfen, and Emmanuel C. Ifeachor. Voice quality prediction models and their application in VoIP networks.
IEEE transactions on multimedia. 2006; 8.4: 809-820.
[14] Y Xie, C Liu, MJ Lee, TN Saadawi. Adaptive multimedia synchronization in a teleconference system. Multimedia
Systems. 1999; 7.4: 326-337.
[15] Gibbon, John F., and Thomas D. C. Little. The use of network delay estimation for multimedia data retrieval. IEEE
Journal on Selected Areas in Communications. 1996; 14.7: 1376-1387.
[16] Shallwani, Aziz, and Peter Kabal. An adaptive playout algorithm with delay spike detection for real-time VoIP.
Electrical and Computer Engineering. IEEE CCECE 2003. Canadian Conference on. 2003; Vol. 2.
[17] Moon Sue B., Jim Kurose, and Don Towsley. Packet audio playout delay adjustment: performance bounds and
algorithms. Multimedia systems. 1998; 6(1):Pp. 17-28.
[18] M.K.Ranganathan and L.Kilmartin. Neural Fuzzy Computation Techniques for Playout Delay Adaptation in VoIP
Networks. IEEE Transactions on Neural Networks. 2005; 16(5).
[19] CJ Sreenan, JC Chen, P Agrawal. Delay reduction techniques for playout buffering. IEEE Trans. Multimedia. 2000;
vol. 2: pp. 88–100.
[20] Chandran P, Lingam C. Performance evaluation of voice transmission in Wi-Fi networks using R-factor. In 2015
International Conference on Information Processing (ICIP), IEEE. 2015; pp. 481-484.
[21] ITU-T, Coded-Speech Database, Supplement 23 to ITU-T P-Series Recommendations, International
Telecommunication Union, 1998.
[22] ITU-T Rec G.107. The E-model: a computational model for use in transmission planning. 2014.

A Statistical Approach to Adaptive Playout Scheduling in Voice Over Internet Protocol Communication

More Related Content

What's hot (20)

Similar to A Statistical Approach to Adaptive Playout Scheduling in Voice Over Internet Protocol Communication (20)

More from IJECEIAES (20)

Recently uploaded (20)

A Statistical Approach to Adaptive Playout Scheduling in Voice Over Internet Protocol Communication