You are on page 1of 17

AMR Voice Quality Improvement Based on PLVA Feature

2016-09-17 – Muhammad Mujtaba


Contents

– Overview & Functionality


– Parameters
– Affected Counters, KPIs and Field Test Elements
– Trial Area and Implementation Plan
– Performance Results and Field Test
– Conclusion and Parameter Settings Scenarios
– Backup Slides

Copyright© 2014 MTN Irancell. All rights reserved


Overview & Functionality

1 Overview
AMR audio codec is a patented speech coding scheme adopted by the 3GPP as the standard
codec for UMTS. Channel encoding for AMR voice services uses convolutional codes. The
Viterbi algorithm is used by most vendors and previous Huawei products. In RAN14.0,
Huawei introduces the AMR Voice Quality Improvement Based on PLVA feature to decode
convolutional codes. The PLVA algorithm outperforms the Viterbi algorithm. Because both
voice and signaling use convolutional codes for channel encoding, the PLVA algorithm can
improve the voice service quality without affecting power control or compromising system
capacity. The feature introduces more robustness to the voice service which is more noticeable
in poor radio condition, and improves the uplink coverage for signaling on the DCH.
The AMR Voice Quality Improvement Based on PLVA feature does not depend on any other
feature and can be used with any other feature.

Copyright© 2014 MTN Irancell. All rights reserved


Overview & Functionality

2 Technical Description
2.1 AMR Speech Encoding and Decoding
AMR is a speech coding standard widely used in the GSM and UMTS communications systems. At the UMTS
physical layer, convolutional codes are used to perform channel encoding for AMR voice services and power control
is used to ensure AMR voice quality.
Figure 2-1 Channel encoding and power control for UMTS AMR voice services in the uplink
Figure 2-2 Channel encoding for classes A, B, and C of narrowband AMR services

Figure 2-2

Figure 2-1

Copyright© 2014 MTN Irancell. All rights reserved


Overview & Functionality
2.2 AMR Speech Decoding Using the Viterbi
Algorithm
The Viterbi algorithm is adopted by most vendors to decode
convolutional codes.
Figure 2-3 Working mechanism of the Viterbi algorithm
The Viterbi algorithm selects the optimal path based
on the maximum likelihood theory and exports the
data decoded on the optimal path. If the data decoded
on the optimal path fails the CRC, the AMR speech
codec discards the data. As a result, voice quality
deteriorates.
Figure 2-4 Narrowband AMR speech decoding using
the Viterbi algorithm
As shown in Figure 2-4, the three AMR voice classes are
separately decoded by the Viterbi decoder, and the CRC is
performed only on the decoded data of class A. The CRCI
is used as a reference for outer loop power control (OLPC)
and is also sent as a bad frame indication (BFI) to the
AMR speech codec on the CN. If a BFI indicates that a
speech frame is erroneous, the AMR speech codec on the
CN discards the speech frame to prevent noise.

Copyright© 2014 MTN Irancell. All rights reserved


Overview & Functionality
2.3 AMR Speech Decoding Using the PLVA
Huawei PLVA is an enhanced CRC-assisted Viterbi
algorithm.
Figure 2-5 Working mechanism of the PLVA
Instead of selecting only the most optimal path, the PLVA selects the top
N optimal paths and performs CRC on the data decoded on these paths.
The PLVA only exports data that passes the CRC. If the data decoded on
these paths fails the CRC, the PLVA exports the data decoded on the
optimal path, which is the same path selected by the Viterbi algorithm.
The PLVA outperforms the Viterbi algorithm because it chooses the data
decoded on multiple paths, which include the optimal one selected by the
Viterbi algorithm. Therefore, when the data decoded by the Viterbi
algorithm is correct, the data decoded by the PLVA is also correct.
However, when the data decoded by the PLVA is correct, the data decoded by
the Viterbi algorithm is not necessarily correct because there are occasions
when the data decoded on the optimal path is incorrect whereas the data
decoded on other paths selected by the PLVA is correct. In simulations where
the PLVA selects four paths, the signal-to-noise ratio (SNR) is 0.2 to 0.8 dB
better than that produced by the Viterbi algorithm.
Figure 2-6 Narrowband AMR speech decoding using the PLVA

Copyright© 2014 MTN Irancell. All rights reserved


Overview & Functionality

As shown in Figure 2-6, CRC is performed on the AMR voice data of class A and therefore the data can be
decoded using the PLVA. The mean opinion score (MOS) of class A AMR voice data is improved by reducing
the BLER. CRC is not performed on classes B and C AMR voice data and therefore classes B and C AMR
voice data can only be decoded by using the Viterbi algorithm. The PLVA exports and sends the decoded data,
Viterbi CRCI, and PLVA CRCI to the RNC. The Viterbi CRCI is used for OLPC. The PLVA CRCI serves as
the BFI, indicating whether a speech frame is erroneous.
This dual-CRCI mechanism does not affect power control. In addition, the performance of the AMR voice
data of class A is improved, maximizing the speech quality. Note that this feature decreases the uplink BLER
and improves the MOS of AMR voice service by using PLVA, and it has no impact on the downlink.

3 Network Impact
System Capacity
No impact.
Network Performance
This feature increases MOSs of AMR voice services which are more noticeable in areas with poor coverage, and improves
the uplink coverage for signaling on the DCH.

4 Activation Observation
The VS.PLVA.User counter informs operators of the number of UEs using this feature. If the value of VS.PLVA.User
counter is not 0, this feature is effective.

Copyright© 2014 MTN Irancell. All rights reserved


Overview & Functionality

5 Performance Monitoring
The gain produced by this feature is reflected by a noticeable increase in the MOS. Take 12.2 kbit/s AMR voice services as an
example. In the uplink simulations, if the BLER is 1%, the MOS is increased by about 0.08; if the BLER is greater than 10%,
the MOS is increased by about 0.35. (The BLER increase is generally caused by UE power limitation, fast channel change, or
strong interference.) In addition, the MOS increase is generally the same under different channel fading conditions.
Figure 5-1 describes the MOS comparison for AMR 12.2k voice (Simulation Result in the uplink).
Figure 5-1 MOS comparison for AMR 12.2k voice (simulation result in the uplink)
Fixed-point test in the lab and drive test are available, which are described as follows:
•Fixed-point test in the lab
To conduct a fixed-point test in the lab, perform the following steps:
1.Deactivate this feature and calculate the MOS.
2.Under the same conditions, reactivate this feature and calculate the MOS in the same way.
The MOS gain is obtained by the average value calculated in step 2 minus that calculated in step 1. The MOS gain is about
0.2.
•Drive test
To conduct a drive test in the lab, perform the following steps:
1.Determine a test route, deactivate this feature, and calculate the MOS.
2.Under the same conditions, reactivate this feature and calculate the
MOS in the same route.
The MOS gain is obtained by the average value calculated in step 2
minus that calculated in step 1. The MOS gain is over 0.1.

Copyright© 2014 MTN Irancell. All rights reserved


Parameters

License Control
Feature ID Feature Name NE Sales Unit
Item

AMR Voice Quality AMR Voice Quality


WRFD-140201 Improvement Based Improvement Based NodeB Per cell
on PLVA on PLVA (Per Cell)

• After the license is activated, this feature has been activated.

Copyright© 2014 MTN Irancell. All rights reserved


Affected Counters, KPIs and Field Test Elements

Copyright© 2014 MTN Irancell. All rights reserved


Trial Area and Implementation Plan

This feature will be tried on HARNCH01 and RRNCH02

Copyright© 2014 MTN Irancell. All rights reserved


Performance Results and Field Test

Summary ::
•Average MOS increase by 0.02 value for both RNC’s- from 4.045 to 4.060 & from 4.047 to
4.062 for HARNCH01 & RRRNCH02 respectively.
•The Percentage sample in Excellent range increase by 3%  From 88.31% to 91.37% & from
88.83% to 91.76% for HARNCH01 & RRRNCH02 respectively.

(5, 4) = Excellent
(4, 3) = Good
(3, 2) = Accept
(2, 1) = Poor
(1, 0) = Bad

Copyright© 2014 MTN Irancell. All rights reserved


Performance Results and Field Test

Copyright© 2014 MTN Irancell. All rights reserved


Performance Results and Field Test

Copyright© 2014 MTN Irancell. All rights reserved


Conclusion and Parameter Settings Scenarios

The MOS improves with the feature.

Copyright© 2014 MTN Irancell. All rights reserved


Backup Slides

Copyright© 2014 MTN Irancell. All rights reserved

You might also like