T Rec G.1051 202303 I!!pdf e
T Rec G.1051 202303 I!!pdf e
T Rec G.1051 202303 I!!pdf e
Recommendation
ITU-T G.1051 (03/2023)
Summary
An important aspect of the data transmission performance of networks are data transfer times and
resulting answering delay in real-time, interactive scenarios. Latency and reactivity are becoming even
more essential for new interactive and real-time applications as e.g., in augmented reality but also in
Industry 4.0 or automotive use.
Latency and resulting reactivity must be measured in a scenario that emulates the application and use
case to be evaluated. This requires first a data transfer profile (traffic pattern) that is considered as
equivalent to the application so that the relevant latency and reactivity can be measured. Second, the
resulting influence of latency to a certain application can be described by an interactivity scoring
model. This model is not a general one, rather, it is individually scaled for each of the use cases like
e.g., e-Gaming or real-time drone control and is focused on scoring transport with a simplified,
parametrizable model approach, it does not target individual application behaviours.
History
Edition Recommendation Approval Study Group Unique ID*
1.0 ITU-T G.1051 2023-03-01 12 11.1002/1000/15468
Keywords
Data traffic patterns, interactivity, latency, round-trip time.
* To access the Recommendation, type the URL http://handle.itu.int/ in the address field of your web
browser, followed by the Recommendation's unique ID. For example, http://handle.itu.int/11.1002/1000/11
830-en.
NOTE
In this Recommendation, the expression "Administration" is used for conciseness to indicate both a
telecommunication administration and a recognized operating agency.
Compliance with this Recommendation is voluntary. However, the Recommendation may contain certain
mandatory provisions (to ensure, e.g., interoperability or applicability) and compliance with the
Recommendation is achieved when all of these mandatory provisions are met. The words "shall" or some other
obligatory language such as "must" and the negative equivalents are used to express requirements. The use of
such words does not suggest that compliance with the Recommendation is required of any party.
© ITU 2023
All rights reserved. No part of this publication may be reproduced, by any means whatsoever, without the prior
written permission of ITU.
1 Scope
The scope of this Recommendation is to specify a method to measure continuously data two -way
latency and loss for a defined observation period. The basis for this method can be IETF's Two-Way
Active Measurement Protocol (TWAMP). The typical symmetrical fixed-rate stream method will be
modified to reflect situations in today's telecommunication networks including mobile scenarios and
typical interactive use cases. The measurement approach is designed to also cover 5G URLLC
configurations.
In this approach, a scalable UDP packet stream is sent from and reflected by a far-end server back to
the measurement client, e.g., from a smartphone or modem device to a server in the network or a
second device.
Based on the results of the latency, latency variation and loss measurements, a generic approach for
a model describing interactivity as a single figure of merit is developed. Because the measurement
results and the interactivity model depend on the chosen traffic pattern emulating an application, there
will be clear rules to derive traffic patterns and corresponding model configurations as well as defined
traffic patterns and formulas for popular applications (by applying the defined rules). This allows an
immediate application of the measurement approach and enables later extension for new, arising
applications.
2 References
The following ITU-T Recommendations and other references contain provisions which, through
reference in this text, constitute provisions of this Recommendation. At the time of publication, the
editions indicated were valid. All Recommendations and other references are subject to revision;
users of this Recommendation are therefore encouraged to investigate the possibility of applying the
most recent edition of the Recommendations and other references listed below. A list of the currently
valid ITU-T Recommendations is regularly published. The reference to a document within this
Recommendation does not give it, as a stand-alone document, the status of a Recommendation.
[ITU-T Y.1540] Recommendation ITU-T Y.1540 (2019), Internet protocol data
communication service – IP packet transfer and availability performance
parameters.
[3GPP TS 23.501] 3GPP TS 23.501 (2020), System architecture for the 5G system (5GS).
[IETF RFC 5357] IETF RFC 5357 (2008), A two-way active measurement protocol (TWAMP).
[IETF RFC 5481] IETF RFC 5481 (2009), Packet delay variation applicability statement.
[IETF RFC 6038] IETF RFC 6038 (2010), Two-way active measurement protocol (TWAMP)
reflect octets and symmetrical size features.
3 Definitions
None.
5 Conventions
None.
6 Introduction
This Recommendation describes the technical realization of two-way delay measurements, how to
apply this approach to emulate real application traffic and how to obtain the metrics and results from
this measurement.
The introduced measurement method considers specific characteristics of mobile networks and, in
general, dynamically adjusted and load-dependent networks and connections.
This Recommendation also gives advice on how to derive load and traffic patterns from real
applications as well as for adjusting the scalable interactivity model to the target use case or
application.
One realization of an on-the-top model to predict interactivity of the tested connection is described in
Annex A.
1 The implementation is not part of [IETF RFC 6038] and is left to the implementor. Preferably, the
implementation carries the information of how many octets should be reflected in each packet. Thus, the
asymmetry of the traffic can be defined per packet as highest granularity.
2 A typical example application is UHD video streaming, where single frames are packetized but exceed the
MTU size and are split. Nevertheless, the entire packet (video frame) is considered as lost if the transport
of a single MTU fails.
3 Please note that IPDV is also used as an abbreviation for IP-packet delay variation in other contexts.
4 Please note that usually the IP kernel of the operating system discards corrupted packets below the IP
interface level. When tracing packets at IP level, corrupted packets are seen as lost.
5 Considering packets exceeding the delay budget as unusable and discarded reflects real-time applications,
where delayed packets cannot be considered for e.g., rendering a media stream.
A.1 Introduction
This annex describes a model for estimation of perceived interactivity. As perceived interactivity
strongly depends on the application class, the computational model must be parametrized for
dedicated application classes and their demands.
The model is based on the obtained two-way latency per packet from simulated data traffic in a
network or transport centric approach.
The basic idea of this model realization is a simple consideration of the application's client and server
and its scalable construction across many different application cases. Even though this will not
perfectly match each individual application, its main advantage is the comparability between
individual settings and modelled test cases due to the same modelling approach being taken.
where a defines the horizontal shift on the t-axis and b the gradient.
For a parametrization that matches the value range of positive latencies and a scaling by f0 to a
maximum score value of fP (0) = fmax, the following formula is applied:
𝑓max 1 1
𝑓P (𝑡, 𝑖) = (1 − 1 ) for t > 0 with 𝑓0 = 1 − 𝑎
𝑓0 − (𝑡 −𝑎)
1+ 𝑒 𝑏 𝑖 1+ 𝑒 𝑏
where t is the packet latency in milliseconds, i is the indicator of the packet, a defines the shift of
fP(t, i) directly on the t-axis in milliseconds and b defines the gradient of fP(t, i), where larger values
of b make fP(t, i) less steep.
The scaling factor f0 guarantees the maximum score value at t = 0. In case, the upper saturation area
of the fP(t, i) starts in the range of t > 0, the f0 will be close to 1.0. If f0 → 1, the parameter a defines
the latency value where the perceived interactivity has fallen to almost 50% of fmax.
In cases where per-packet two-way latency can be obtained, the logistic function can be applied to
each individual packet latency. Here, the value fP(t, i) is computed for each individual packet i and its
latency t. The aggregated interactivity based on latencies IL for an observation period is the average
of all fP(t, i):
1 𝑓𝑚𝑎𝑥 1
𝐼𝐿 = ∑𝑁
𝑖=1 (1 − 1 )
𝑁 𝑓0 − (𝑡 −𝑎)
1+ 𝑒 𝑏 𝑖
In cases where the per-packet two-way latency cannot be obtained in the test set-up or higher order
statistics appear more applicable, the principle of the logistic weighting can be applied to the median
of the two-way latencies as a single value input into the function.
If seen from an quality of experience (QoE) perspective, first, parameter a shifts the latency value,
where the decrease of perceived interactivity depending on latency starts along the t-axis. In case f0
is close to 1.0, parameter a directly defines the latency where the predicted interactivity is decreased
to 50% of the maximum reachable score. Second, b determines the sensitivity of the user, means the
range of latency, where the perceived interactivity is decreasing from almost maximum to close to
zero. Both parameters are depending on the expectation of the user to the application used.
Figure A.2 – Parametrization of the logistic weighting function for two-way latency
A.3 Consideration of lost and discarded packets and packet delay variation in the model
A.3.1 Disqualified packets
The defined packet loss PL reports packets which are not received and cannot be used by the
application. In addition to that, a real application will also discard packets which are heavily delayed
6 This separation into IPDV per link is also required in case of asymmetrical traffic patterns using different
packet frequencies in each direction. There is no one-to-one relation between sent and received packets
which is the pre-condition for a two-way PDV or two-way IPDV.
Figure A.3 – Example of influence of PDVsQ and PDQ to the logistic weighting function
Generally, the above-mentioned formula is applied to the entire observation period (e.g., 10 s). To
result in a higher temporal granularity, the formula can also be applied to shorter sub-intervals of 1 s,
for example.
A.4 Conclusion
The described computational formula is a simplified, generic approach to retrieve perceived
interactivity from latency measurements with the advantage of transparent scalability. The logistic
weighting function to estimate a perceived interactivity value from two-way latency measurements
ensures a monotonous behaviour and can model perceptual saturation areas at both boundaries of the
scale similar to typical perceptual weightings. Degrading effects of delay jitter (derived from PDVsQ)
and disqualified packet loss (PDQ) are considered by simple multiplicative scaling factors.
Multiplicative scaling retains the monotonous behaviour and does not change the saturation areas and
their cut-off positions.
For parametrization of the interactivity model as described in clause 9.2.2, the following principles
are used: A fluent video stream produces 60 frames per second (fps), a movie 24. It is assumed that
degradation starts if the channel adds a two-way delay of ~30 ms (two frames delay for 60 fps).
Furthermore, a degradation to 60 of 100 for the interactivity score is assumed in case of a two-way
delay added by the channel of ~60 ms (four frames for 60 fps). These thresholds can be seen as
challenging but refer to highly interactive gaming applications in high speed networks as in 5G
URLLC.
In addition, it is anticipated that a PDVsQ of 30 ms reduces the interactivity as defined by latency by
another 10% (DPDV = 0.9) and a ratio of disqualified packets of 5% reduces the perceived interactivity
by 20% (DDQ = 0.8).
The parametrization of the model:
𝐼𝑛𝑡𝐴𝑐𝑡 = 𝐼L × 𝐷PDV × 𝐷DQ
1 𝑓max 1
with DPDV = 1 – 𝑃𝐷𝑉sQ / u, DDQ = 1 – v PDQ and 𝐼L = ∑𝑁
𝑖=1 (1 − 1 )
𝑁 𝑓0 − (𝑡 −𝑎)
1+ 𝑒 𝑏 𝑖
is resulting in:
Parameter fmax a b u v
Parameter fmax a B u v
Value 100 22 6 40 20
The chosen packet size is 1000 bytes, which is typical for video content. The packet sending
frequency is then derived to be 62 packets/s during the sustainable phases and 250 packets/s in the
high bit rate phases. An example for a 'video chat HD' application is a similar use case, but it relies
on live camera video and transmitting the participant videos in high resolution and the main focus is
visual interaction and feedback between people. According to real video chat applications, there is a
short initial (set-up) phase followed by a sustainable phase where the video link is established.
The chosen packet size is again 1000 bytes, which is typical for video content. The packet sending
frequency is then derived to be 62 packets/s during the sustainable phases and 125 packets/s in the
high bit rate phases. The traffic pattern is symmetrical, meaning the packet size and frequency are the
same for uplink and downlink.
The applicable delay budget is identical for both types of video application. In [3GPP TS 23.501], a
maximal one-way latency for conversational video of 150 ms is defined (5QI class 2), the two-way
latency therefore should not exceed 300 ms and this value forms the delay budget for this test case.
Parameter fmax a b u v
Value 100 215 50 150 30
II.1 Introduction
Subjective tests show that subjects are less sensitive to delay and delay jitter in a real gaming situation
than the 3GPP thresholds define. There can be many reasons, but in subjective testing the actual
application on server and client are considered.
Even though the model described in Annex A is based on measurements on the transport layer and
the measurements do not consider the application itself, the model in Annex A can be parametrized
to approximate the subjective perception as in subjective tests.
The given examples in this Appendix II are parametrized by a real gaming situation, while the
parameters in Appendix I are based on QoS limits. The subjects played a CS_GO FPS game of
90 seconds match rounds over a Steam Link connection. Consequently, the example parameters in
this appendix differ from those given in Appendix I, because the model approximate subjective scores
include a real application and its gaming situation.
As already mentioned above, dividing the packet pattern into temporal sub-segments of 40 ms is
expected to be accurate enough. However, to handle the different number of packets per link direction
the protocol of choice needs to support, or be extended to support, the concept of bursts (in this case
frames) instead (or in addition to) of the single packet concept. This results in two-way measurements
such as RTT on the frame burst level instead of the single packet level, but only for the first packet
on the respective link.
The target value used for the function parameterization as in Annex A for the cloud gaming over
mobile networks use case is selected to be the answer to the question regarding the overall QoE
[b-ITU-T G.1072]. The decision for this approach is based on the outcome of an interactive test for
Finally, the effect of packet loss is modified compared to Annex A, because the 3GPP RTT specified
limit (RTT = 100 ms) does not reflect when interaction with the service (cloud gaming in this case)
is not possible per se, since packets are rarely late for a critical moment, and users adapt to latency
which anyway is handled by 𝐼L.
The proposed modification is to consider a limit of 100 ms for disqualification on large jitter (IPDV)
values in order to capture the fact that large delay spikes disrupt the ability to interact. The reasoning
for this is that tests showed that delay spikes start to affect at 50 ms and can have very noticeable
effect at 1500 ms, and these jitter spikes are not captured by the median value of PDV.
In addition, it is proposed that DQ term is used to describe the packet loss. Therefore, the 𝐷DQ term
becomes:
𝐷DQ−CG = 1 − 𝑣 × 𝑃DQ
𝑃𝐽𝑆
with 𝑃DQ = + 𝑃L−CS + 𝑃L−SC
𝑇MeasureWindow
Parameter a b u v
Value 213 91 25 7
It should be noted that foffset is set to 1.0 and fmax to 4.0 to scale the output into a five-point ACR scale
as in the underlying subjective test.
Series E Overall network operation, telephone service, service operation and human factors
Series F Non-telephone telecommunication services
Series G Transmission systems and media, digital systems and networks
Series H Audiovisual and multimedia systems
Series I Integrated services digital network
Series J Cable networks and transmission of television, sound programme and other multimedia
signals
Series K Protection against interference
Series L Environment and ICTs, climate change, e-waste, energy efficiency; construction, installation
and protection of cables and other elements of outside plant
Series M Telecommunication management, including TMN and network maintenance
Series N Maintenance: international sound programme and television transmission circuits
Series O Specifications of measuring equipment
Published in Switzerland
Geneva, 2023