Page 212 - ITU Kaleidoscope 2016

2016 ITU Kaleidoscope Academic Conference

network conditions. The limitations of traditional linear and formulate of the proposed method. Section 4 deals with
regressive models need to be rectified where the seasonality the algorithm development procedures. The implementation
of a regularly repeated pattern could be eliminated with environment along with different metric of performance is
increased focus towards accurate predictions. In this paper, a described in Section 5. The result and discussion is presented
new algorithm based on a non-linear stochastic model called, in Section 6. Section 7 concludes with scope for future work.
Auto Regressive Integrated Moving Average (ARIMA)
Based Bit Rate Adaptation (ABBA) is proposed. The ABBA 2. SYSTEM ARCHITECTURE
algorithm is modeled on a time series consisting of sequence
of sampled bit rate over a continuous time interval to analyze 2.1 The Client Server Model
a successive statistical measurement that has no natural
ordering for the observations. This stochastic model is used The proposed system architecture emulates the client server
for trend analysis of the forthcoming bit rates, which decides model where server’s job gets simplified on the expense of
the resolution of video to be sent from the server. The client’s increased monitoring and analysis process. The
strategic decisions are based on the successive observation server consists of three sub modules: i. frame capture, ii.
of sampled bit rate in a regular time interval to understand streaming the video, and iii. receiving feedback. On the other
the nature of series. hand, the client consists of three modules: i. decoder / player,
To analyze the performance of the proposed ABBA ii. stream flow analysis, and iii. receiver’s feedback. The
algorithm, two existing approaches: i. Heuristic Decision video being streamed is encoded dynamically using ITU-T
Rate Adaptation (HDR), and ii. Buffer Switching Rate H.264 [12] video codec. The live (or stored) video is
(BSR) algorithms has been formulated and developed here. streamed from the server to client through 4G wireless
The HDR [7] employs the difference in arrival time of networks and the client scrutinizes the link bandwidth and
packets and buffering time as inputs for predicting the near analyses its trend to make an intelligent decision based on
future using a set of decision rules whereas the BSR [8] prevailing scenario. This decision is sent as a feedback to
monitors the buffer occupancy level dynamically and server which tries to match channel capacity and the sends
chooses the mode of operation based on its fill percentile the video at corresponding resolution and frames per second.
using harmonic mean to effectively identify the nature of The client samples the incoming bit rate and monitors the
network for streaming the videos. pattern with an aim to analyze and predicts the near future
The ABBA, HDR, and BSR algorithm were implemented bandwidth. The server is notified with the predicted link
using VLC Framework in Java (VLCJ) that is completely capacity which in turn responds with content adaptation
open source and can easily be plugged into to the existing process. The bit rates of packets are related to a time series
systems. The system level implementation was in adherence model where a set of data points denotes the bit rates over
to the ITU-T J.247 recommendation (Table 1) that describes successive time. The sampled data are arranged in a proper
about the ‘objective perceptual multimedia video quality chronological order continuously and the past observations
measurement’. The developed system were tested for delay are analyzed to develop a mathematical model (ABBA
variability ITU-Y.1540 [9] and quality of video were algorithm) that captures the underlying data to make
observed using PSNR ITU-R J.340 [10] and VQM ITU- strategic decisions. This parametric approach considers that
J.149 [11] along with other standard popular video quality the underlying stochastic process has a certain structure
evaluation metrics. which can be described using two parameters: auto-
correlations and auto-covariance to forecast the future bit
Table 1. Test factors as per ITU guidelines rates using regression.

Parameter Standard Metrics
2.1.1 Sender Sub-modules
Frame Rate 5 to 30 fps

Codec H.264
The server’s main job is to acquire the media content live
Resolution QCIF,CIF,VGA
ITU –T from the camera or fetch from a memory location in case of
Temporal errors J.247 <=2 sec
a stored video. The following modules represent the
QCIF 16 kbps to 32kbps workflow of streaming at the server side (Fig. 2).
Min bandwidth CIF:64 kbps to 2Mbps
Required a) Init: This module initializes the VLCJ player and
VGA 128 kbps to 4 Mbps
ITU-R J.340 PSNR >=25 identifies the media locator required for transmission.
Delay Variation b) Stream: It is used to establish connection with the
ITU-T (Quantile and min delay requesting client using sockets while creating instance of
Performance Y.1540 difference should not be >50
Metrics player to stream at required quality.
ms)
ITU-T J. VQM [0-1] c) Adapt: This module receives message from the client and
149 uses this feedback to adapt to the network prevailing
conditions by interpreting the data from client to make a
strategic decision.
The rest of the paper is organized as follows. The client-
server model of the system architecture is presented in
Section 2. Section 3 describe about the mathematical model

– 194 –