Page 73 - Kaleidoscope Academic Conference Proceedings 2021

Connecting physical and virtual worlds

VR2-1, Total Packet Size PDF VR2-1, Total Packet Size PDF
6000 6000
Data Sample Data Sample
Fitting
5000 Fitting 5000
4000 4000
Count hits 3000 Inherent Characteristic: Count hits 3000 Statistical Characteristic:
4
Mean = 5.9023 10 (Byte)
4
Mean = 5.992 10 (Byte)
Standard deviation = 4481 (Byte)
2000 2000 Standard deviation = 2265 (Byte)
1000 1000
0 0
2 3 4 5 6 7 8 2 3 4 5 6 7 8
PacketSize (Byte) 10 4 PacketSize (Byte) 10 4
(a) Alt1:Using inherent characteristic (b) Alt2:Using statistical characteristic

Figure 1 – Dual eye fitting results comparison

including the mean and variance. In Gaussian fitting, as shown in the figures. The accuracy of the Alt
the fitting function can be expressed as 2 traffic modeling has been acknowledged by quite
some companies during RAN1#104b-e meeting, based
2
(x − b 1 )
f(x) = a 1 exp − , (1) on which the traffic model generated accordingly has
c 2
1 been agreed for evaluation wherein the maximum and
minimum packet size is generated according to the 3σ
where a 1 , b 1 and c 1 are the output of the estimation.
principle of Gaussian distribution [9]. Moreover, the
The non-normalized PDF of Gaussian distribution
method can be extended to other XR traffic modeling
equation is copied as below:
including traffic arrival jitter modeling and the file
2 size/arrival jittering of multiple streams.
(x − µ)
f(x) = a exp − , (2)
2σ 2
2.2 Multi-streams model
where a denotes the amplitude of PDF of Gaussian
distribution, µ denotes the mean value of Gaussian An XR service typically consists of multiple flows with
distribution and σ denotes the standard deviation of different Quality of Service (QoS) requirements ranging
Gaussian distribution. Comparing Eq. (1) and Eq. (2), differently in terms of data rate/reliability/latency etc.
the following two correspondence can be observed: Differentiation in data rate can be translated in a file
size ratio belonging to different flows taking into account
a 1 = a, (3)
the traffic arrival rate. Based on SA study [12][13],
multi-streams should be representative principally in
the following two use cases: the traffic of I/P frame
and of tile-based field of view (FoV)/non-FoV model.
b 1 = µ, (4)
For the I-P frame model, according to the data sample
provided by SA4 [12][13], ratios of the file size can be
visible as in Table 1. The file size ratio α 1 for I-frame
√ and P-frame can thus be set to α 1 = 2 reasonably.
c 1 = 2σ. (5)
For the FoV/non-FoV multi-streams model, a tile-based
At last, we can generate the PDF curve based structure for FoV streaming is shown as in Figure
√
on estimated b 1 (mean value) and c 1 / 2 (standard 2. According to [14], the FoV stream arrives in a
deviation). Fine-tuning of the values around b 1 and pattern of 30-tile group, each of which contains 18
√
c 1 / 2 can still be done according to step 3 in Alt 2 tiles arriving in a second, while that of a non-FoV
to cater to the deficiency of M and N set up in step 1. stream is video streaming at 30 frames per second.
For the same VR2-1 traffic, the fitting result using Alt2 Scattered in the streaming window, the transmission of
is shown in Figure 1(b). these slim tiles does not mandate bulky and sometimes
overbooked radio resource reservation; this can be quite
Compared with the mean/variance generation using
advantageous in the face of coexistence of other services
the raw sample statistics (Alt 1), the traffic derived
[3].
with an additional pre-filtering process (Alt 2) is much
more in close vicinity to the actual sample distribution

– 11 –