Академический Документы
Профессиональный Документы
Культура Документы
March 2015
SwissQual AG
Allmendweg 8 CH-4528 Zuchwil Switzerland
t +41 32 686 65 65 f +41 32 686 65 66 e info@swissqual.com
www.swissqual.com
No part of this publication may be copied, distributed, transmitted, transcribed, stored in a retrieval system, or translated
into any human or computer language without the prior written permission of SwissQual AG.
Confidential materials.
All information in this document is regarded as commercial valuable, protected and privileged intellectual property, and is
provided under the terms of existing Non-Disclosure Agreements or as commercial-in-confidence material.
When you refer to a SwissQual technology or product, you must acknowledge the respective text or logo trademark
somewhere in your text.
SwissQual®, Seven.Five®, SQuad®, QualiPoc®, NetQual®, VQuad®, Diversity® as well as the following logos are
registered trademarks of SwissQual AG.
Diversity Explorer™, Diversity Ranger™, Diversity Unattended™, NiNA+™, NiNA™, NQAgent™, NQComm™, NQDI™,
NQTM™, NQView™, NQWeb™, QPControl™, QPView™, QualiPoc Freerider™, QualiPoc iQ™, QualiPoc Mobile™,
QualiPoc Static™, QualiWatch-M™, QualiWatch-S™, SystemInspector™, TestManager™, VMon™, VQuad-HD™ are
trademarks of SwissQual AG.
SwissQual acknowledges the following trademarks for company names and products:
Adobe®, Adobe Acrobat®, and Adobe Postscript® are trademarks of Adobe Systems Incorporated.
Intel®, Intel Itanium®, Intel Pentium®, and Intel Xeon™ are trademarks or registered trademarks of Intel Corporation.
Microsoft®, Microsoft Windows®, Microsoft Windows NT®, and Windows Vista® are either registered trademarks or
trademarks of Microsoft Corporation in the United States and/or other countries U.S.
Contents
Benchmarking of VoLTE Services A first field experience ......................................................................... 0
4 Conclusion .........................................................................................................................................14
CONFIDENTIAL MATERIALS ii
Benchmarking of VoLTE Services
Figures
Figure 1: Basic flow in a VoLTE to VoLTE connection ...................................................................................... 2
Figure 2: Example of SIP message flow in a VoLTE to VoLTE call .................................................................. 3
Figure 3: Call Set-Up times for mobile to PSTN connections ............................................................................ 4
Figure 4: Call Set-Up times for mobile to mobile connections........................................................................... 4
Figure 5: Voice codec information in VoLTE RTP header ................................................................................. 5
Figure 6: IP Throughput in a VoLTE call ........................................................................................................... 5
Figure 7: Examples for listening quality for VoLTE in comparison to 3G calls .................................................. 6
Figure 8: Examples of quality distribution for VoLTE and 3G calls ................................................................... 7
Figure 9: Examples of audio delay in mobile to PSTN calls .............................................................................. 9
Figure 10: Examples of audio delay in VoLTE and 3G mobile to mobile calls .................................................. 9
Figure 11: Basic principle of a jitter buffer to compensate packet delay jitter ................................................. 10
Figure 12: Example of an aligned pair of reference and degraded signal ....................................................... 11
Figure 13: Example of variable delay in a VoLTE live test sample ................................................................. 12
Figure 14: Example of variable delay in a 3G live test sample........................................................................ 12
Figure 15: Occurrence of delay changes for VoLTE and 3G live network samples ........................................ 13
Figure 16: P.863 MOS-LQO statistics in relation to delay changes in the test sample ................................... 13
Foreword
In this white paper we take an in depth look at the performance of recently launched commercial VoLTE
networks by evaluating speech quality. The information presented is based on real world data collected from
an end-to-end perspective – simulating the end user experience. The data used was collected between
June of 2014 and January 2015 on three different VoLTE operators in the US. This white paper does not
cover core network architecture and call handling within the network. What is covered is the attributes of a
VoLTE network and how they impact voice quality. Test methodology is also covered in detail.
This document will provide first results and methodologies of testing and benchmarking VoLTE. It treats
VoLTE from an end-to-end measurement perspective. The focus of the document revolves around the typical
measurements made in all voice telecommunications networks.
In general, the meaning of KPIs and metrics are the same for VoLTE as for legacy technologies. This is
required if operators have to benchmark a VoLTE service with a legacy voice call in 2G / 3G under
comparable conditions. When we look deeper into the technical details that enable a VoLTE call to function,
it is fairly straightforward when we evaluate a VoLTE to VoLTE call within the same IMS Server. Beyond this
simple case, however, complexities quickly arise. The following call types need special attention to ensure
KPI are not compromised.
- Subscriber is in LTE, having VoLTE service but calling a 2G/3G/PSTN subscriber
- Subscriber is in LTE, having an established VoLTE call but loosing LTE coverage
(SRVCC, Single Radio Voice Call Continuity)
- Subscriber is in LTE, having no VoLTE service, just LTE data access
(CSFB, Circuit-Switched Fall-Back)
The following sections provide some real world examples for analysis. These measurement results were
taken from live VoLTE networks using commercially available devices. These numbers illustrate the
important parameters associated with running a quality VoLTE network. They can be used for setting a
benchmark upon which to improve, or to evaluate one carrier vs. another.
1
It has to be noted that LTE capable devices are in LTE in idle mode and fall back to 3G/2G in the event of a voice call.
This CSFB called strategy extends the call setup in 3G/2G by around 1s compared to phones that have no LTE support
at all and CSFB is not applied.
accounts for the small difference (3.74 vs 3.66) in voice quality that the VoLTE based calls have over the 3G
calls if the coding is the same?
The explanation can be found in the distribution of the MOS values shown in Figure 8. For both, the
maximum quality is determined by the used compression technology that is AMR-WB 12.65kbps. The
maximum value recorded for AMR-WB falls in the range of a MOS value of 4.0 to 4.2. Note there is no
technology dependent disadvantage with VoLTE or 3G.
2
There are rare cases, where in handovers one frame gets lost or duplicated, but it results in a delay variation of one
voice frame (20ms).
Figure 10: Examples of audio delay in VoLTE and 3G mobile to mobile calls
As expected, a mobile to mobile connection in circuit-switched 3G with common transcoding shows the
highest audio delay. There are two air-links but two compression / de-compression steps applied. The shown
average of 300ms is a fair value, but there are many examples of networks with more than 400ms of audio
delay – barely acceptable. Almost identical same values in audio delay were measured when calling a
VoLTE device from a 3G device.
In both cases there is one operator ‘C’ who applies Transcoding-Free Operation that enables a much shorter
transmission time because one coding/decoding step is avoided. This operator ‘C’ is even processing 3G to
VoLTE calls without transcoding and keeps the audio wideband capability across the technologies! This will
not only result in a short audio delay but also significantly increased audio quality.
When looking at VoLTE to VoLTE, the audio delay is much shorter than in a normal 3G call; however it
results in about the same delay as measured with a 3G to 3G call using AMR-WB TrFo.
The given figures are real field examples of well implemented technology and designed devices. In practice
there might be delays much higher than these, a delay measurement may help to optimize towards those
almost perfect values.
Figure 11: Basic principle of a jitter buffer to compensate packet delay jitter
In case of very long packet delay or packet loss, it may happen that the buffer runs empty (buffer underrun).
In this case, the decoder has no more information to decode. Clever strategies are used to solve this issue.
Intelligent decoders recognize an up-coming under-run and try to extend speech pauses to empty the buffer
slower in the hope that new packets are received in the meantime. If there are no speech pauses, the
speech information itself gets stretched, either by time-warping or by creating additional voice frames similar
to the replacement strategies in voice decoders. All these smoothing strategies try to gain time by bridging
long delays of the next expected packet. However, these strategies can just bridge a limited time when the
buffer runs empty. The worst case is that the smoothing strategies reach their own limits and voice is inter-
rupted due to excessive buffer underrun. The decoder simply has nothing to work with. It is logical that all
these issues are less common with the use of a larger jitter buffer. Conversely, these issues will be more
common if the jitter buffer is short. There is a direct relation to the overall audio delay as discussed in the
previous section. A long jitter buffer can avoid warping the voice signal but it has to pay with a higher audio
delay in general.
However, each strategy for stretching the voice stream or even pausing accumulates more and more delay
in the signal. Once packets are received again and the buffer gets filled, decoders try to get rid of the
accumulated delay again. Usually, speech pauses get shortened for that.
Effects on the voice are very different. Extending or shortening speech pauses are almost imperceptible.
They do not change the speech signal itself. It is different if the voice signal itself gets stretched or
compressed, which influences the talk spurts, they are getting longer or shorter. A pure extension or
compression as a re-sampling will change the spectral distribution too, since the voice pitch gets higher or
lower and may sound unnatural. Therefore, time-warping with pitch preservation is used. This way a
moderate stretching or compression of the voice becomes much more acceptable.
In consequence, packet delay jitter widely becomes compensated fully; uncompensated delays can be – to
some extent – masked by smart voice processing. Finally, uncompensated packet delay jitter affects the
temporal structure of the voice stream and can be measured physically, but its effects on speech quality
depend on the used processing strategy in the decoder and can only be evaluated with a speech quality
algorithm.
The following measurements in real VoLTE and 3G calls show the range of delay variations in 6s talk spurts,
it means the difference from the minimum to the maximum of measured short-term delays.
The following detailed pictures illustrate the relative delay during a transmitted speech sample in a VoLTE
channel (Figure 13). It can be seen that the delay increases (stepwise) during the first sentence and gets
adjusted at the end of the speech pause again.
Figure 15: Occurrence of delay changes for VoLTE and 3G live network samples
In 3G almost all samples (94%) have no delay changes, the threshold of 20ms corresponds to one AMR
frame and there are only 6% measurements above that value. For VoLTE the picture is different, about 70%
of the samples show delay changes of >20ms.
But does this variable delay matter from the user perspective in terms of audio quality? This analysis cannot
be done one to one, since the test samples are also affected by other live distortions too. Especially for 3G,
delay changes are usually combined with handovers and potential gaps and short muted speech parts which
also have an obvious effect on MOS values.
However, an indication of speech quality degradation can be derived if the MOS scores are analysed
separately for different amounts of variable delay. The diagram in Figure 16 shows the P.863 averages for
samples with a dedicated range of delay changes vs. time warping.
Figure 16: P.863 MOS-LQO statistics in relation to delay changes in the test sample
For VoLTE with variable delay the situation is quite stable. Even if we can measure a variable delay
physically in the signal, it hardly decreases the speech quality. This underlines that the jitter compensation
strategies as focusing on speech pauses and moderate stretching with pitch-preservation can maintain a
good audio quality.
4 Conclusion
Voice over LTE is becoming today’s reality in more and more live networks. Within this transition period a
careful benchmarking of the VoLTE services itself as well as of legacy technologies is extremely important to
guarantee a quality standard that is comparable to well-known 2G and 3G services. For quite a while both
technologies will be used in parallel. It is not only important to benchmark the technologies to each other but
also to benchmark cross-technology connections. This is the daily experience for the majority of VoLTE
users’ calls and in the near future VoLTE to CS interconnections will be the majority of mobile calls in
general.
This transition period may last the next few years and requires a dense supervision by benchmarking and
optimization teams. Measurements on the application layer using POLQA help to guarantee the best voice
call experience. As can be seen in the data presented above, optimizing certain elements such as jitter
buffers can help improve conversation quality and when done correctly, it can do this without a terrible
compromise to audio quality. This report gives first example values for voice call performance and highlights
the trade-off between speech quality and audio delay. The results indicate that VoLTE may provide the best
voice call performance overall, appearing to be better than today’s 3G networks.
However, the integration of VoLTE functionality into handsets plays an important role in the end user voice
experience, much more than in the past. Benchmarking systems used to test VoLTE networks must use
commercially available devices designed for use on the network they are testing. It is also critical for the
measurement system to emulate real world voice call use cases along with having the full complement of RF
(Layer 1, 2, and 3), SIP, IMS, and RTP, and all other TCP/IP measurement capabilities to really understand
what is behind good or bad network performance. This will ensure that network operators will have a true
image about the user’s satisfaction when using VoLTE.
The collected and analyzed results in this study show already impressive and rapid improvements made
within the months from early commercial launch. The data collection was made in a period where one
operator has already changed to transcoding-free AMR-WB processing across technologies while others
have still to do this step. The recognition of this advantage could be done on the application layer by
measuring high quality in wide audio bandwidth in combination with a short transmission delay. Both are
valid indicators that this technology is enabled and will live up to expectations.
Conversely, the road ahead to a 100% IP based communications world will be full of pitfalls. The upcoming
change over from 2G/3G systems to all IP based LTE is happening now and with that there will be more re-
farming of spectrum, continued explosion of LTE bands deployed, carrier aggregation, WiFi offloading, etc.
etc. As network operators strive to deliver all the promises of high speed data networks, it will be critical for
them to keep their eye on voice performance. Nothing could be more damaging than customer churn
created by a poor VoLTE experience that is only consuming 12kpbs of data bandwidth.
Chapter 4 | Conclusion 14
CONFIDENTIAL MATERIALS