Professional Documents
Culture Documents
43.
The information element VSQI (Video Streaming Quality Index) estimates the
viewer-perceived video and audio quality achieved during video streaming
sessions. This chapter takes a look the VSQI algorithm.
See also the document Video Streaming Quality Measurement with VSQI
which is included in the TEMS Investigation documentation package.
Compare chapter 44 on MTQI.
43.1.
Like SQI (chapter 41), VSQI is a no-reference method which judges the
quality of the received signal on its own merits, without knowledge of the
original.
The kind of subjective test which VSQI strives to imitate is one where viewers
are instructed to assess both video and audio and combine their perception
of each into an overall multimedia quality score.
The output from the VSQI algorithm is expressed as a value between 1
and 5, conforming to the MOS (Mean Opinion Score) scale which is
frequently used in subjective quality tests. The unit for VSQI is called MOSVSQI.
43.2.
UM-457
43.3.
VSQI does not directly consider the signal presented to the human viewer;
that is, no analysis of perceptual input is performed to detect specific visible
artifacts. The transferred video is not analyzed frame by frame in any way.
Thanks to the monitoring of packet loss (item no. 2 in section 43.2 above),
however, even slight problems with blockiness, jitter, and so on will still be
noticed by the algorithm and affect the VSQI score.
43.4.
Two versions of the VSQI algorithm have been devised: one static and one
dynamic version.
Static VSQI is presented in the event Streaming Quality VSQI. It does not
appear as an information element. Dynamic VSQI, on the other hand, is
contained in the information element Streaming VSQI.
43.4.1.
Static VSQI
The static version of VSQI takes an entire streamed video clip as input and
assigns a single quality score to it.
Input parameters to the static version of VSQI are as follows:
UM-458
Here, VSQIclean is the clean value obtained for the clip prior to transmission.
This score is determined by the quality of the encoding, which is in turn
dependent on the choice of codecs and bit rate.
The size of the buffering penalty depends on the time taken for initial
buffering, the time spent rebuffering, and the number of rebuffering events.
The size of the packet loss penalty is determined as follows. A running packet
loss average over the last 4 s is computed approximately every second, and
the values thus obtained are weighted and summed to yield an appropriate
overall measure of the packet loss. The latter is then translated into a
deduction from the VSQI score.
The static VSQI algorithm has been fine-tuned for clips of around 30 s and
should therefore in practical use be applied to clips of similar duration. The
video sequences must not be too short because of how the buffering works:
each instance of rebuffering takes several seconds to complete, and
moreover if the clip is short enough it will have been buffered in its entirety
before the replay starts, so that no rebuffering will ever occur. For clips
considerably longer than 30 s, on the other hand, disturbances towards the
end will be more harshly penalized by viewers than those occurring early on,
simply because the late ones are remembered more vividly. Therefore, since
the current VSQI algorithm does not take into account such memory effects, it
would probably perform slightly worse for long clips. (The dynamic version of
VSQI naturally is not affected by this limitation.)
43.4.2.
UM-459
The graph below shows in rough outline the different ways in which dynamic
VSQI can evolve during the replay of a streaming video clip. The best
achievable quality, i.e. the ceiling in the graph, is dependent on the codec/bit
rate combination but is also affected by the amount of packet loss. In this
example the packet loss is assumed to be constant so that the influence of
buffering can be clearly discerned.
1 The user tolerates (and might even expect) a certain amount of initial
delay; but the longer the buffering drags on, the more the user loses
patience.
2 Once the replay gets going, the perceived quality picks up again and
soon approaches the highest achievable level.
3 If rebuffering occurs, VSQI deteriorates rapidly. Rebuffering events are
much less tolerated by viewers than initial buffering, especially if
repeated; VSQI captures the latter by making the slope of the curve
steeper for each new rebuffering event.
4 After the replay has recommenced, VSQI recovers reasonably quickly,
but not infrequently from a rock bottom level.
UM-460
44.
MTQI (Mobile TV Quality Index) is a refinement of the video part of the VSQI
quality measure (see chapter 43). Unlike VSQI, MTQI does not include an
assessment of audio quality.
The MTQI algorithm can be concisely characterized as follows.
Algorithm components:
UM-461
45.
This chapter deals with certain aspects of VQmon video and audio perceptual
quality metrics, which are computed in the course of HTTP streaming. The
metrics are listed in Information Elements and Events, section 3.8: Streaming
IEs; for the service testing setup, see section 12.16.6.4 of the present
document.
The VQmon algorithms have been developed by Telchemy, and the
information that follows is taken from Telchemy documentation.
45.1.
VQmon provides a set of Mean Opinion Scores (MOS) estimating the quality
of each video and audio stream as perceived by end-users. Each MOS value
ranges from 1 to 5, where 1 represents the verdict Unacceptable and 5
means Excellent:
MOS-V: Video MOS, considering the effects of the video codec, frame
rate, packet loss distribution, and group-of-pictures structure on video
quality.
MOS-A: Audio MOS, considering the effects of the audio codec, bit rate,
sample rate, and packet loss on viewing quality.
45.1.1.
UM-462
For example, a video viewed on a handset with a small screen might receive
an absolute MOS of 3.1 when little or no quality degradation is evident, while
for a handset with a larger screen, the same MOS value might suggest that
some noticeable impairments were present.
To facilitate quality comparisons between different video service types,
VQmon provides both Absolute and Relative MOS-V:
45.2.
The VQmon metric Video Service Transmission Quality (VSTQ) is a codecindependent score that measures the ability of the network to reliably
transport video. VSTQ is expressed in the range 050.
UM-463