ps kpi
TRANSCRIPT
-
7/27/2019 PS KPI
1/38
GSM BSS Network PS KPI (Download Rate) Optimization Manual INTERNAL
Product Name Confidentiality Level
GSM BSS INTERNAL
Product Version
Total 38 page
GSM BSS Network PS KPI (Download Rate)
Optimization Manual
(For internal use only)
Prepared by GSM&UMTS Network PerformanceResearch Department
Date2008-02-29
Reviewed by Date
Reviewed by Date
Granted by Date
Huawei Technologies Co., Ltd.
All rights reserved
-
7/27/2019 PS KPI
2/38
GSM BSS Network PS KPI (Download Rate) Optimization Manual INTERNAL
Revision Record
Date Version Description Author
2008-02-29 1.0 Draft completed Wang Guanghua
2008-11-01 1.1 Reorganized Geng Haijian
(ID: 00105443)
2008-12-17 1.1 Revised according to review comments Geng Haijian
(ID: 00105443)
2008-11-01 All rights reserved Page 2 of 38
-
7/27/2019 PS KPI
3/38
INTERNAL
Contents
1.1 Position of the GPRS or EGPRS Network in End-to-End Applications.............................................................. ....8
1.2 Introduction to CQT and DT....................................................................................................................................9
1.3 Performance Baseline of the (E)GPRS Network....................................................................................................10
1.4 Introduction to Channel Resource Management Algorithm...................................................................................11
1.5 Introduction to Link Quality Management Algorithm............................................................................................14
1.6 Mechanism of Link Synchronization/Channel Synchronization............................................................................18
2.1 Downloading Large-Sized Files in CQT in Idle Hours to Identify the Problem....................................................20
2.1.1 Failure to Assign or Steadily Occupy Four Channels .....................................................................................22
2.1.2 Failure to Steadily Occupy a High Coding Scheme ........................................................................................23
2.1.3 Error Block.......................................................................................................................................................24
2.1.4 High Ratio of Control Blocks...........................................................................................................................25
2.1.5 Abnormal Release of TBF......................................................................................................................... .......30
2.1.6 High Transmission Rate at the RLC Layer But Low Transmission Rate at the Application Layer.............. ...30
2.2 Competition of the Download Rate through CQT..................................................................................................31
2.3 Comparision between DT downloading rates in busy and idle hours....................................................................31
3.1 Problem Identification............................................................................................................................................34
3.2 Routine Optimization..............................................................................................................................................37
4.1 Appendix 1: Parameter Description........................................................................................................................38
4.2 Appendix 2: Optimization Manual for Parameters on Server and Test PC.................................................... ........38
2008-11-01 All rights reserved Page 3 of 38
-
7/27/2019 PS KPI
4/38
INTERNAL
Figures
GPRS networking.............................................................................9
Payload of the GPRS protocol stack ................................................10
Channel allocation example............................................................12
Information shown by the TEMS log when you download large-sizedfiles in ideal situation.....................................................................21
Messages traced at the application layer when you download large-sized files in ideal conditions..........................................................22
Timeslots assigned in Packet Downlink Assignment message (on theleft) and Packet Timeslot Reconfigure message (on the right) .........23
Viewing the Receive block bitmap (URBB) in the Packet DownlinkACK/NACK message to determine the error blocks ..........................24
PDP context of the MS....................................................................26
MS flow control data reported by the PCU to the SGSN.....................27
Application layer data traced at the MS side....................................28
Checking whether TBF is abnormally released through the ACKmessage........................................................................................30
Comparison of the radio transmission environment before(upper) andafter (lower) the network swapping................................................32
Frequent reselections caused by improper neighboring cellconfiguration and reselection parameter settings............................33
2008-11-01 All rights reserved Page 4 of 38
-
7/27/2019 PS KPI
5/38
-
7/27/2019 PS KPI
6/38
INTERNAL
2008-11-01 All rights reserved Page 6 of 38
-
7/27/2019 PS KPI
7/38
INTERNAL
References
[1] EDGE DT Download Rate Optimization Thoughts and Cases, at http://support.huawei.com, released on June26, 2008
[2] TCP/IP Detail Volume 1, by W. Richard Stevens, in April, 2008
[3] GPRS Network Technology, by Motorola Engineering Institute, on June 1, 2005
2008-11-01 All rights reserved Page 7 of 38
-
7/27/2019 PS KPI
8/38
INTERNAL
GSM BSS Network PS KPI (Download Rate)
Optimization Manual
1 OverviewThe first two chapters of this document describe the download process and performance
analysis. Readers who are concerned only about the GPRS network optimization can start
from chapter3.
The fault identifying tools in this document mainly refer to TEMS and Ethereal/Wireshark.
These tools can be used to trace or browse the information at the NE side.
The description in this document is based on the condition that the TDM transmission is usedover the Abis interface and the FR transmission is used over the Gb interface.
1.1 Position of the GPRS or EGPRS Network inEnd-to-End Applications
For end-to-end applications, an MS functions like the network card of the client to connect to
the GGSN through the GPRS network, and then to the Internet. This process is the same asthat when accessing the Internet through cables. The difference is that the MS is not
connected to the router directly through cables but through the GPRS network. Compared
with cables of 100 Mbit/s or higher bandwidth, the GPRS network has a longer RTT delay
and smaller bandwidth. In addition, the delay and the bandwidth vary with the actualconditions.
From the perspective of the download rate, the target of GPRS network optimization is to
increase the bandwidth and reduce the delay (the advantages of a small delay are quite
obvious when you download small-sized files).
2008-11-01 All rights reserved Page 8 of 38
-
7/27/2019 PS KPI
9/38
INTERNAL
Figure 1 shows the GPRS networking.
Figure 1 GPRS networking
Figure 1 shows a typical GPRS networking mode based on E1 transmission. Each interfacesupports direct connection. Timeslot cross devices, that is, DXX devices, can be used for
transmission on the Gb interface and the G-Abis interface. If the external PCU is installed, the
PCU provides the Gb interface. The Pb interface between the PCU and the BSC usuallyadopts the E1 direct connection mode.
1.2 Introduction to CQT and DT
Telecom operators assess the rate performance of a GPRS network through call quality tests
(CQTs) and drive tests (DTs).
Why do telecom operators choose CQT and DT to assess the network performance?
As a fixed-point test, the CQT is conducted in places where the wireless conditions are goodand the C/I ratio is steady. The CQT performed in idle hours help check all the network
elements and transmission links between the Um interface and the Gi interface. Such CQT
can absolutely show the performance of the equipment. The CQT performed in busy hourshelp check the quality of the resource management algorithms, including the algorithms
related to channels, Abis interface resources, and Gb interface resources.
The CQT performed in busy hours, however, bring great uncertainty. For example, if anotheruser is using the download service during the test, the download rate will be greatly affected.
In this case, the CQT cannot fully reflect the performance of the equipment. This is becausethe impact of resource allocation is great. The test result can be used only for comparison of
the performance before and after the network swapping.
The DT result may be affected by large C/I fluctuation and cell reselection. The DT can be
used to assess the wireless coverage and interference, the quality of the algorithm foradjusting the coding scheme, and the processing performance of the PCU during cell
reselection (as the handover function in the PS domain has not achieved yet). The DT,
however, also bring uncertainty. For example, in the case of red light, whether the C/I ratio is
good or and whether the signals are in deep fading points have great impact on the average
rate in the DT.
2008-11-01 All rights reserved Page 9 of 38
-
7/27/2019 PS KPI
10/38
INTERNAL
The algorithm for adjusting the coding scheme is also referred to as the link quality control algorithm, orIR/LA algorithm. The reason for adjustment is that the required wireless quality, that is, C/I ratio, varies
with the coding schemes. For a specific C/I ratio, a proper coding scheme should be employed, to helpachieve the optimal balance between the data amount sent per unit time and the retransmission rate, thus
maximizing the transmission rate.
1.3 Performance Baseline of the (E)GPRSNetwork
Calculate the theoretical performance limit of the product in the case that resources aresufficient.
Assume that the MSS value of the TCP negotiated between the server and the MS is 1460.This is the default size of a TCP/IP data packet sent on the Ethernet. The data packet is
encapsulated according to the Ethernet protocols and a 20-byte TCP header and a 20-byte IPheader are added. The MTU value of the intermediate network is 1500. That is, no
fragmentation is performed. The size of an LLC PDU negotiated between the MS and the
SGNN is 506. The MSC9 coding scheme is employed on the Um interface. Figure 2 shows
how the data is encapsulated.
Figure 2 Payload of the GPRS protocol stack
Actually, if the MSS value is 1450 bytes, the data can be fragmented into three SNDCP packets to
achieve the highest encapsulation efficiency. When the MSS value is 1460 bytes, the efficiency to theLLC layer is calculated as follows: 1460/ (1460+20+20+13+24) = 94.99%. When the MSS value is 1450
bytes, the efficiency to the LLC layer is calculated as follows: 1450/(1450+20+20+10+18) = 95.52%.
For an EGPRS network, when the MCS9 is employed, the theoretical rate of a single channel
is 59.2 kbit/s. If four channels are used for transmission, one channel will be used as thecontrol channel. The control information accounts for about 19% of the data on the channel
before the Uplink ACK/NACK optimization scheme in uplink extension mode is
2008-11-01 All rights reserved Page 10 of 38
-
7/27/2019 PS KPI
11/38
INTERNAL
incorporated, and the control information accounts for about2% of the data on the channelafter the preceding optimization scheme is incorporated. Therefore, the data rate is 59.2 x (4-
2%) = 235.616 kbit/s. Multiply the rate by the efficiency of the LLC layer to get the
maximum rate at the application layer in ideal situation, that is, 235.616 kbit/s x95.52%=225.06 kbit/s.
To achieve a rate close to the theoretical limit, do as follows: Download a large-sized file.
During the first stage of download, the TCP connection is just established, and the TCP uses
the slow start mechanism. Slow start means that the TCP layer sends data at a slow rate to
avoid network congestion when the TCP layer does not know the network transmissionbandwidth or quality or when the network transmission bandwidth is reducing or the
transmission quality is degrading. Therefore, the volume of the initially sent data is
insufficient. Each node tries not to discard packets, frames, or blocks. These nodes refer to theIPBB, core network, Gb interface, PCU, G-Abis interface, BTS, and Um interface, as shown
in Figure 1. The flow control at each interface cannot be controlled to the extent that the data
is not enough for sending. The bandwidth of the radio interface must be guaranteed and a
maximum number of channels should be occupied. Currently, the multislot capability of mosttesting MSs is 10 or 11, and a maximum of four downlink timeslots can be occupied. The
channels are not multiplexed by other MSs, and the MCS9 coding scheme can be used
steadily.
1.4 Introduction to Channel ResourceManagement Algorithm
Allocating as many channels as possible to the MS is a method for ensuring the radio
interface bandwidth. The channel resource management algorithm allocates channels based onthe maximum capability of an MS. That is, the number of channels allocated to an MScorresponds to the multislot capability of the MS. In addition, the algorithm balances TBFs
among channels when possible. Block resources are allocated in the following principles: The
bandwidth for GBR users is guaranteed. The best fairness is achieved. That is, the polling
mechanism is used for the TBFs that are multiplexed onto the same channel.
The entire channel resource management algorithm consists of channel allocation, dynamicchannel conversion/release, and load balance. Channel resources consist of the channel pool
of CS (CSD) and the channel pool of PS (PSD). After configuration, static PDCHs are
grouped to the PSD, and dynamic PDCHs are grouped to the CSD. Dynamic channelconversion is to group part of CSD channels to the PSD. The dynamic channel conversion can
be triggered by the following conditions: inadequate multislot capability, EGPRS MSassigned to a GPRS channel, and load exceeding the Uplink Multiplex Threshold of Dynamic
Channel Conversion/Downlink Multiplex Threshold of Dynamic Channel Conversion.Channel allocation is to assign the optimal channel group in the PSD to the MS. Load balance
is to redistribute all TBFs within a timed cycle, to balance the load among channels.
The channel resource management algorithm is complicated. Here the downloading process
of an MS is taken as an example and the configuration is typical. Figure 3 shows the channelallocation process.
Configuration: Maximum Ratio Threshold of PDCHs in a Cell = 100; Uplink Multiplex
Threshold of Dynamic Channel Conversion/Downlink Multiplex Threshold of Dynamic
Channel Conversion = 12/12.
2008-11-01 All rights reserved Page 11 of 38
-
7/27/2019 PS KPI
12/38
INTERNAL
PDCH Uplink Multiplex Threshold/PDCHDownlink Multiplex Threshold = 70/80.
Maximum PDCH numbers of carrier = 8.
Figure 3 Channel allocation example
(1) During initial access (no matter whether the multislot capability of the MS is known), thePSD has only one channel. Thus, the uplink and downlink data are both multiplexed onto this
channel. The requirements of the multislot capability of the MS are not met. The multislot
capability of the MS supports four downlink timeslots, but only one timeslot is available now.In this case, dynamic channels are converted. Three TCHs are converted into PDCHs.
(2) After 4.5 seconds, the load balance flow is triggered, and another channel allocation is
performed on this TBF. The amount of data flow cannot be checked at this time. Therefore,
the service is considered as a neutral service. Three downlink timeslots and two uplink
timeslots are assigned to the MS. Timeslots 5, 6, and 7 are the three downlink timeslotsassigned.
(3) The service is identified as the download service after 4.5 seconds. In this case, timeslots
are assigned to this MS in 4+1 mode. Timeslots 4, 5, 6, and 7 are the downlink timeslots
assigned to the MS.
(4) Another MS accesses the network. The service is judged as neutral at the initial phase.Timeslots 4, 5, and 6 are assigned to the second MS. Timeslot 6 on the uplink is occupied by
the first MS. Therefore, when the second MS accesses the network, try to avoid the
2008-11-01 All rights reserved Page 12 of 38
-
7/27/2019 PS KPI
13/38
INTERNAL
multiplexing of timeslot 6 by the two MSs. Timeslot 5 is assigned to the second MS. Then,timeslots 4, 5, and 6 are assigned for the downlink data transmission of the second MS
according to the multislot capability of the second MS.
(5) It is found that the multislot capability of the second MS is not met, and another channel isrequired. In this case, the channel mapping to timeslot 3 is converted according to the priorityand the timeslot re-assignment flow for the second MS is triggered.
(6) In this case, the total number of TBFs on all channels is eight, and the number of occupied
channels is five. Therefore, the multiplexing degree is 8/5 = 1.6 > 1.2. The heavy load triggers
dynamic channel conversion. The number of converted channels is calculated as follows:
8/1.2 5 + 1 = 2 (Total number of TBFs in the Cell/Threshold of Dynamic Channel
Conversion -Number of PDCHs occupied + 1, where 1 is used as it is an integer). The
requirements of the Maximum Ratio Threshold of PDCHs in a Cell and Maximum PDCH
Numbers of carrier are met. In this case, channels mapping to timeslots 1 and 2 are
converted. When load balance is performed on the TBF, this TBF is assigned to timeslots 1, 2,
3, and 4.
When channel allocation does not satisfy the multislot capability of the MS, consider the
following aspects:
Channel type. If a channel is configured as an EGPRS dedicated channel, it cannot be
assigned to a GPRS MS. If a channel is configured as a dedicated channel for EGPRS,
and the channel is occupied by an EGPRS MS, the channel cannot be assigned to aGPRS MS. If a GPRS MS occupies an EGPRS dedicated channel, the GPRS MS is
moved from the channel.
When the multiplexing of a channel reaches the PDCH Uplink/Downlink Multiplex
Threshold, this channel cannot be assigned.
When an exception occurs on the channel, such as out-of-synchronization, the channelcannot be assigned.
Discontinuous channels cannot be assigned.
IfAllow E Down G Up Switch is set to Close and a downlink TBF of an EGPRS MS is
on this channel, the uplink data of a GPRS MS cannot be multiplexed onto this channel.IfAllow E Down G Up Switch is set to Close and an uplink TBF of a GPRS MS is on
this channel, the downlink data of an EGPRS MS cannot be multiplexed onto this
channel.
When Reassignment TBF for Different Trx is set to Not Allow, no re-assignment will
be performed even if the number of initially assigned timeslots does not meet themultislot capability of the MS.
When PS Concentric Cell HO Strategy is set to No handover between underlaid
subcell and overlaid subcell orHandover from overlaid subcell to underlaid subcell,the channels in the overlaid subcell cannot be assigned.
When the Abis interface timeslot is unavailable, the channel cannot be assigned.
For an external PCU, if no PCICs are available, the channel cannot be assigned.
Constraints of dynamic channel conversion are as follows:
When the value of (number of static PDCHs + converted dynamic channels)/total
number of service channels) reaches the value ofMaximum Ratio Threshold of
2008-11-01 All rights reserved Page 13 of 38
-
7/27/2019 PS KPI
14/38
-
7/27/2019 PS KPI
15/38
INTERNAL
Table 1 Mapping between BEP and coding scheme
CV_BEP
0 1 2 3 4 5 6 7
MEAN_BEP
0 MCS-3 MCS-3 MCS-3 MCS-3 MCS-3 MCS-3 MCS-3 MCS-3
1 MCS-5 MCS-5 MCS-5 MCS-5 MCS-3 MCS-3 MCS-3 MCS-3
2 MCS-5 MCS-5 MCS-5 MCS-5 MCS-3 MCS-3 MCS-3 MCS-3
3 MCS-5 MCS-5 MCS-5 MCS-5 MCS-5 MCS-3 MCS-3 MCS-3
4 MCS-5 MCS-5 MCS-5 MCS-5 MCS-5 MCS-3 MCS-3 MCS-3
5 MCS-6 MCS-5 MCS-5 MCS-5 MCS-5 MCS-5 MCS-5 MCS-5
6 MCS-6 MCS-5 MCS-5 MCS-5 MCS-5 MCS-5 MCS-5 MCS-5
7 MCS-6 MCS-6 MCS-6 MCS-5 MCS-5 MCS-5 MCS-5 MCS-5
8 MCS-6 MCS-6 MCS-6 MCS-5 MCS-5 MCS-5 MCS-5 MCS-5
9 MCS-6 MCS-6 MCS-6 MCS-5 MCS-5 MCS-5 MCS-5 MCS-5
10 MCS-6 MCS-6 MCS-6 MCS-6 MCS-5 MCS-5 MCS-5 MCS-5
11 MCS-6 MCS-6 MCS-6 MCS-6 MCS-6 MCS-5 MCS-5 MCS-6
12 MCS-6 MCS-6 MCS-6 MCS-6 MCS-6 MCS-6 MCS-6 MCS-6
13 MCS-6 MCS-6 MCS-6 MCS-6 MCS-6 MCS-6 MCS-6 MCS-6
14 MCS-7 MCS-6 MCS-6 MCS-6 MCS-6 MCS-6 MCS-6 MCS-6
15 MCS-7 MCS-6 MCS-6 MCS-6 MCS-6 MCS-6 MCS-6 MCS-6
16 MCS-7 MCS-7 MCS-6 MCS-6 MCS-6 MCS-6 MCS-6 MCS-6
17 MCS-7 MCS-7 MCS-7 MCS-7 MCS-7 MCS-7 MCS-7 MCS-7
18 MCS-7 MCS-7 MCS-7 MCS-7 MCS-7 MCS-7 MCS-7 MCS-7
19 MCS-7 MCS-7 MCS-7 MCS-7 MCS-7 MCS-7 MCS-7 MCS-7
20 MCS-7 MCS-7 MCS-7 MCS-7 MCS-7 MCS-7 MCS-7 MCS-7
21 MCS-7 MCS-7 MCS-7 MCS-7 MCS-7 MCS-7 MCS-7 MCS-7
22 MCS-8 MCS-8 MCS-8 MCS-8 MCS-8 MCS-8 MCS-8 MCS-8
23 MCS-8 MCS-8 MCS-8 MCS-8 MCS-8 MCS-8 MCS-8 MCS-8
24 MCS-8 MCS-8 MCS-8 MCS-8 MCS-8 MCS-8 MCS-8 MCS-8
25 MCS-8 MCS-8 MCS-8 MCS-8 MCS-8 MCS-8 MCS-8 MCS-8
26 MCS-8 MCS-8 MCS-8 MCS-8 MCS-8 MCS-8 MCS-8 MCS-9
27 MCS-9 MCS-9 MCS-9 MCS-9 MCS-9 MCS-9 MCS-9 MCS-9
2008-11-01 All rights reserved Page 15 of 38
-
7/27/2019 PS KPI
16/38
INTERNAL
CV_BEP
0 1 2 3 4 5 6 7
MEAN_BEP
28 MCS-9 MCS-9 MCS-9 MCS-9 MCS-9 MCS-9 MCS-9 MCS-9
29 MCS-9 MCS-9 MCS-9 MCS-9 MCS-9 MCS-9 MCS-9 MCS-9
30 MCS-9 MCS-9 MCS-9 MCS-9 MCS-9 MCS-9 MCS-9 MCS-9
31 MCS-9 MCS-9 MCS-9 MCS-9 MCS-9 MCS-9 MCS-9 MCS-9
Table 1 describes how to select a coding scheme for new blocks. For the GPRS network, thecoding scheme that is used to send the retransmitted data blocks is the same as that used to
transmit the new blocks. For the EGPRS network, a lower coding scheme is adopted to send
the retransmitted data blocks, and this coding scheme depends on the following three factors:
the coding scheme for sending the earlier data block, coding scheme that is the same as thatfor sending new blocks, and link quality management mode (IR or LA). Select one-cluster
coding schemes. For example, MCS3, MCS6, and MCS9 belong to one cluster. The extent of
lowering is determined by IR or LA. For example, if the LA is adopted, MCS9 must belowered to MCS3. If the IR is adopted, MCS6 is enough. In IR mode, the MS buffers part of
the data of the block that is not decoded just now and then softly-combine and decode it
according to the retransmitted data. The data can be decoded in this way. Therefore, it is
recommended that you adopt the IR for the test. But the soft combination function of certainMSs is not good, or some MSs do not support such function at alll; in this case, it is possible
that the IR function brings a negative gain.
The coding scheme is selected on the basis of the following conditions:
1. Wireless environment. Table 2 shows the quality of wireless signals required by the coding
schemes.
Table 2 Requirements of different coding schemes for the Um interface
Coding Scheme Required ReceivingLevel of the MS(dBm)
Required C/I Ratio inTU3 Mode (dB)
MCS1 -102 13
MCS2 -101 15
MCS3 -99 16.5
MCS4 -97 19
MCS5 -98 18
MCS6 -96 20
MCS7 -93 23.5
2008-11-01 All rights reserved Page 16 of 38
-
7/27/2019 PS KPI
17/38
INTERNAL
MCS8 -90.5 28.5
MCS9 -86 30
2. Transmission quality of the G-Abis interface
If out-of- synchronization occurs on the sublink, a high coding rate cannot be used.
Data blocks encoded according to high coding schemes do not carry the synchronization
header. When serious code slipping occurs, the network sides chooses to use a lowcoding scheme with the synchronization header instead.
3. Number of timeslots on the Abis interface. Table 1 describes the number of 16 kbit/sAbis interface timeslots required by the coding schemes.
Table 1 Requirements of different coding schemes for the timeslots at the Abis interface
EGPRS GPRS Number of Required16 kbit/s Timeslots onthe Abis Interface
MCS1-MCS2 CS1-CS2 1
MCS3-MCS6 CS3-CS4 2
MCS7 3
MCS8-MCS9 4
The timeslot on the Abis interface can serve as: a statistically multiplexed signaling link (oneOML for each BTS; one RSL for each TRX); a voice channel (one 16 kbit/s timeslot is
required for each channel); a PDCH (several 16 kbit/s timeslots are required for each channel.
One is called the main link, and other links are sublinks.)
In the case that the Flex Abis function is not enabled, the voice channel and PDCH is bound toone 16 kbit/s timeslot during timeslot configuration. For the MSs that are performing the
packet service, if the coding scheme should be adjusted and the Abis resources are required,
apply for idle timeslots on the Abis interface. If the application is approved, adjust the coding
scheme. When the coding scheme is lowered, the release of Abis resources will not betriggered. When the channel is idle, the Abis resources are released when Timer of ReleasingAbis Timeslot times out.
In the case that the Flex Abis function is enabled, the Abis timelot is not bound to a channel
during channel configuration. An Abis interface timeslot is applied and bound when thechannel is activated. That is, the original main link is also grouped into the idle Abis resource
pool. Based on the principle that the CS service is preferred, a channel should be
preferentially assigned to the CS service. If no Abis timeslot is available, the channel that is
earlierly assigned to the packet service will be preempted.
Therefore, when idle timeslots are insufficient:
2008-11-01 All rights reserved Page 17 of 38
-
7/27/2019 PS KPI
18/38
INTERNAL
Configure all the remaining timeslots as idle timeslots in the case that the Flex Abis
function is disabled. Check whether the customer uses DXX devices for timeslot cross. If
DXX devices are used, the number of available timeslots is not certainly the number oftimeslots on the E1. In this case, ask the customer about the number of available
timeslots
Calculate the signaling link multiplexing degree according to the traffic model.Increasing the multiplexing degree can conserve Abis interface resources
Enable the Flex Abis function to help realize sharing of the Abis interface resources
Abis interface resources of Huawei products are divided into resource pools in the unit of BTS, and allcells of a BTS share the Abis resources. In addition, for cascaded BTSs, part of the Abis interfacetimeslot resources of the upper-level BTSs should be reserved for the lower-level BTSs.
You can set the MaxAbisTSOccupied parameter to specify the maximum number of Abis interface
timeslots that can be bound on a carrier.
1.6 Mechanism of LinkSynchronization/Channel Synchronization
Each PDCH is bound to several Abis interface timeslots. One of the timeslots is the main link,
and others are sublinks. At the initial phase after the channel is converted, channelsynchronization occurs. This process takes about one second. The PCU sends a
synchronization frame through the main link to the BTS. The BTS returns a synchronization
frame to the PCU. Thus, the differences of frame number and block number between the two
frames can be calculated. Thus, the PCU sends data packets based on the advance. When asynchronization frame is sent, if no synchronization header is found, or the check fails after a
synchronization frame is received, the channels cannot be synchronized.
Each channel contains several Abis interface links. The synchronization should be performed
separately on each link. Each subsequent frame sent on the link contains a synchronizationheader (the header is not carried when MCS6 or MCS7 is used). The main link also sends
signaling frames in CS1, and the sublinks send idle information frames. These frames also
contain synchronization headers. Therefore, certain changes to the synchronization headers
are acceptable. But if the synchronization headers change once in every ten minutes, the PCUwill restrict the frames in high coding scheme.
If synchronization of the main link fails, synchronization of the channel fails. If
synchronization of a sublink fails, the coding scheme will be adjusted. When the transmission
quality on the G-Abis interface is decreased to a certain level, synchronization of the linkfails. The transmission quality on the G-Abis interface is reflected by the frame error rate on
the G-Abis interface. The frame error rate on the G-Abis interface is calculated through the
following formula: frame error rate on the G-Abis interface = (Number of Received Check
Error TRAU Frames + Number of Received Out-of-Synchronization TRAU Frames)/(Number of Sent Valid TRAU Frames + Number of Sent Empty TRAU Frames).
If the frame error rate is less than 10e-5, the link is of good quality, and the MS can transmit
data steadily
2008-11-01 All rights reserved Page 18 of 38
-
7/27/2019 PS KPI
19/38
INTERNAL
If the frame error rate is between 10e-5 and 10e-4, it has certain impact on the data rate theMSs in transmission state.
If the FER is higher than 10e-4, the link is quite unstable and tends to be out of
synchronization. In this case, an MS can hardly transmit a large amount of data.
Possible causes of a high frame error rate or loss of synchronization on a link on the G-Abis
interface are as follows: loss of synchronization of the E1 clock with the Um interface clockof the BTS and unsteady transmission quality. You can check whether the high frame error
rate or loss of synchronization is caused by transmission problems by performing local or
remote loopback at the BTS side. You can also connect the test tools to both ends to capture
data packets for analysis.
2008-11-01 All rights reserved Page 19 of 38
-
7/27/2019 PS KPI
20/38
INTERNAL
2 Thoughts About IdentifyingDownload Rate Problems
As described in section 1.2, compared with the CQT, the DT involves changes to the coding
scheme and cell reselection due to C/I fluctuation. Actually, the link quality control algorithmcan achieve balance between coding scheme and retransmission rate, and changing the coding
scheme changes the actual bandwidth on the Um interface. The impact of the slow start
process when the TCP connection is just established is on downloading small-sized files is
more obvious than that on large-sized files. Therefore, when the download rate is low, youshould perform tests of downloading large-sized files in radio transmission environments
where the C/I ratio is good, so as to check whether the transmission on each node is normal.
2.1 Downloading Large-Sized Files in CQT inIdle Hours to Identify the Problem
Figure 4 shows the rate recorded in the TEMS log when you download large-sized files in idlehours in good radio transmission environment.
2008-11-01 All rights reserved Page 20 of 38
-
7/27/2019 PS KPI
21/38
INTERNAL
Figure 4 Information shown by the TEMS log when you download large-sized files inideal situation
Dummy control block: The PCU dispatches a block every 20 ms. The sending sequence is as follows:NACK block>VS block>PACK block. When the PCU has no such blocks to send, the PCU sends
the Dummy control block. The TEMS counts Dummy control blocks as control blocks. That is, thecontrol blocks counted by the TEMS consist of real control message blocks and Dummy control blocks.
The messages traced at the application layer of the MS shows that the delay of each block isfixed. The MS returns an ACK message after receiving every two TCP packets without packetloss or disorder. The strict test results of the download process should contain only the
interaction messages with the FTP server. No interaction messages with other IP addresses
should be included. Figure 5 shows the messages traced at the application layer of the MS.
2008-11-01 All rights reserved Page 21 of 38
-
7/27/2019 PS KPI
22/38
INTERNAL
Figure 5 Messages traced at the application layer when you download large-sized filesin ideal conditions
The following sections specify how to identify the problems based on the messages traced at
the application layer and the TEMS log files.
2.1.1 Failure to Assign or Steadily Occupy FourChannels
You can use several methods to check the number of assigned channels through the TEMS.You can check the number of yellow grids shown in Figure 4. Another method is to view the
value displayed in GSM data timeslot. These two methods, however, cannot show theaccurate number of assigned channels. To obtain the accurate value, locate the latest PacketDownlink Assignment orPacket Timeslot reconfiguration message, as shown in Figure 6.
2008-11-01 All rights reserved Page 22 of 38
-
7/27/2019 PS KPI
23/38
INTERNAL
Figure 6 Timeslots assigned in Packet Downlink Assignment message (on the left) andPacket Timeslot Reconfigure message (on the right)
According to the description in section 1.3, for the downloading task performed in idle hours,
TS 1, TS 3, and then TS 4 may be assigned. Normally, multiple downlink timeslots can be
assigned to the MS. However, if a channel cannot be occupied, check the channelconfiguration to see whether the PDCHs, consisting of static and dynamic channels, aresufficient. If the PDCHs are insufficient, you should configure enough PDCHs on the BCCH
TRX. Then, check whether loss of synchronization occurs on the channel through the alarms
related to the PCU. For the external PCU, run the mt pdch show state allcommand to check the status of all PDCHs in this cell. For a built-in PCU, run the DSP
PDCH command to check the channel status.
The failure to occupy four channels, including preemption of the channel by the voice service,
does not occur in tests in idle hours. Otherwise, the failure is caused by a channel fault.
2.1.2 Failure to Steadily Occupy a High CodingScheme
This situation is classified into the following scenarios:
Long occupation of a low coding scheme. Possible causes are as follows: fixed downlinkcoding scheme and inadequate Abis interface resources. If the downlink coding scheme
is fixed, change the configuration to non-fixed coding scheme. If the Abis interface
resources are insufficient, increase the number of available Abis interface timeslots
according to the description in section 1.5.
Untimely adjustment of the coding scheme. If downward adjustment is untimely, the
block error rate is high. In this case, you can solve the problem by reducing the value of
2008-11-01 All rights reserved Page 23 of 38
-
7/27/2019 PS KPI
24/38
INTERNAL
BEP period. If upward adjustment is untimely, another possible cause is that the bindingof the Abis interface timeslot takes a long time. For the test, if the initial adjustment is
slow, the Downlink Default MCS Type can be set to MCS9.
2.1.3 Error BlockError block means that the MS does not receive the data block sent by the PCU. There is no
standard to determine an acceptable block error rate. For different channel models, the blockerror rates are different in the same configuration.
How to check the block error rate? As shown in Figure 4, BLER/TS(%) is the block error
rate calculated by the TEMS on the basis of the total number of received blocks. To check
whether a specific data block is received, view the Packet Downlink ACK/NACK message, as
shown in Figure 7.
Figure 7 Viewing the Receive block bitmap (URBB) in the Packet Downlink
ACK/NACK message to determine the error blocks
As shown in Figure 7, the analysis of the Packet Downlink ACK/NAC message in the EGPRS
network by the TEMS is incorrect. For the GPRS network, the analysis of NACKed Blocknumbers by the TEMS is correct.
The Receive block bitmap (URBB) shows the error blocks. Why these blocks are not
received? If the data block loss is not caused by the Um interface factors or G-Abis interface
factors, check whether the failure of receiving blocks is caused by the receiving of the system
information from the neighboring cell before the receiving of the Packet DownlinkACK/NACK message.
The possible causes of the failure to receive data blocks are as follows:
1. A high bit error rate occurs on the Um interface, and the error bits are sequential.
Consequently, data block loss occurs. The test is performed in good radio transmission
2008-11-01 All rights reserved Page 24 of 38
-
7/27/2019 PS KPI
25/38
INTERNAL
environment. Therefore, the high block error rate is not caused by the Um interfacefactors. In addition, in the DT, if the block error rate is high, you can modify the values
ofBEP Period, BEP Filter and BLER Filter.
2. The data block loss is caused by the error frame or out-of-synchronization frame on the
G-Abis interface.
3. The MS is busy and thus cannot receive data blocks. As stipulated by the protocol, the
MS must decode the data on the BCCH on the TRX serving one of the six neighboring
cells with the strongest signal level within 30 seconds. When the signal fluctuation
occurs and a large number of neighboring cells is configured, the MS may frequentlyresolve the system information of the neighboring cells. In this case, you need to cut
down the number of neighboring cells to avoid unnecessary neighboring cells.
2.1.4 High Ratio of Control Blocks
In Huawei products, the MS is assigned with only one control channel. The control channel is
bidirectional. Thus, the uplink and downlink data is multiplexed onto the same controlchannel. In this way, you can locate the control channel. As shown in Figure 4, TS 4 is the
control channel. That is, real control messages are sent only through TS 4. Before the uplink
ACK optimization scheme in uplink extension mode is incorporated, the proportion of the
control messages is 17% to 21%. After the uplink ACK optimization scheme in uplinkextension mode is incorporated, the proportion of the control messages is about 2%. If a non-
control channel transmits control blocks, which are actually Dummy blocks, or the proportion
of control blocks on the control channel is large, it indicates that the PCU has no data to
transmit, so the PCU transmits the Dummy control blocks instead.
The sequence in which the PCU sends data is as follows: NACK blocks (the Packet Downlink
ACK/NACK message through which the MS confirms the error blocks); VS blocks (new
blocks, the RLC fragments the LLC PDUs according to the bytes born by different coding
schemes); PACK blocks (blocks that are not confirmed by the MS). If the preceding datablocks are not available, the PCU sends Dummy control blocks. Note that VS blocks are sent
only when the PCU sending window is not full.
The possible causes of the high ratio of control blocks may be as follows:
1. The subscribed peak rate is not high. As shown in the GSM PDP context in Figure 8, thepeak rate is 128000 octets/s =128000 x 8/1024 kbit/s = 1000 kbit/s, which is greater than
the theoretical rate limit.
2008-11-01 All rights reserved Page 25 of 38
-
7/27/2019 PS KPI
26/38
INTERNAL
Figure 8 PDP context of the MS
2. The RLC adopts the acknowledged mode. In this case, the I-frame mechanism is used
for data sending. The next frame is sent only when the peer end acknowledges the
receiving of the previous frame. In acknowledged LLC mode, connection at the LLClayer should be established and released. This inevitably increases the signaling traffic at
the LLC layer. In a word, the acknowledged LLC mode greatly affects the download
rate. You can check the operation mode of the LLC layer through the PDP context of the
MS. If the acknowledged LLC mode is used, you need to change the mode to theunacknowledged mode at the SGSN side. The subscription information of the SIM card
also needs to be modified.
3. The data sending window stalls. This problem usually occurs in GPRS networks,
because GRPS networks support only a window with 64 data blocks. Regardless of the
cause of error blocks, the RRBP delay of Huawei products is about 200 ms. When an MSreports the receiving of error blocks, 200 ms already passed. If the MS occupies fourtimeslots at this time, the PCU already sends out 200 ms/(20 ms/block) x 4 = 40 blocks,
which easily cause the window to stall. Two methods can be used to identify this
problem. You can increase the RRBP frequency to see whether the error block rate isreduced. Or you can measure the amount of received data on the Gb interface in a certain
period to check whether the amount of received data is larger than the data sent by the
PCU.
4. Improper flow control on the Gb interface. Check the Flow Control MS message traced
on the Gb interface to see whether the reported value is smaller than the data sending rate
of the PCU. If the reported value is smaller than the data sending rate of the PCU, itmeans that the flow control is improper. The flow control data on the Gb interface is
shown as follows.
2008-11-01 All rights reserved Page 26 of 38
-
7/27/2019 PS KPI
27/38
INTERNAL
Figure 9 MS flow control data reported by the PCU to the SGSN
Why flow control is performed on the Gb interface? A GPRS network uses a long narrow transmission
channel. The sever keeps sending packets till the window of the server is stalled. The sent packets mustbe buffered in the GPRS network, but the memory size of the PCU is limited and cannot buffer all the
packets. In this case, the SGSN needs to share part of the packet load while ensuring that the PCU hasenough data to send. Therefore, the protocol defines that the PCU sends flow control parameters,
including the parameters specifying the buffer size and data sending rate, to the SGSN to control thedata flow. The SGSN determines each packet to be sent according to the reported flow control
parameters. Assume that the PCU sends packets according to the reported rate. That is, the data in thebuffer decreases according to the rate. This amount of data decreased in the PCU buffer in the interval
between two packets sent by the SGSN can be calculated. If the SGSN sends the packet to the PCU, the
amount of the packet is added to the amount of data in the PCU buffer. If the amount of data exceeds thebuffer size of the PCU, the SGSN does not send the packet.
The mechanism of flow control for Huawei products is as follows: A flow control message is reported
when the TC timer expires. If the amount of the data in the buffer is larger than 90% or lower than 10%,a flow control message is also reported. If the SGSN does not respond to the flow control message, the
flow control message will be re-sent after the FC timer expires.
5. Insufficient bandwidth on the Gb interface
When the SGSN is connected to the PCU in the FR transmission mode, the load is
shared by multiple NSVCs. Each NSVC is carried on the BC. The BC consists of
multiple 64 kbit/s timeslots on the E1 link. Thus, the physical bandwidth can be
calculated, that is, number of the timeslots x 64 kbit/s.
It is recommended that you calculate the Gb bandwidth by measuring the Downlinkdata kbytes sent to FR per NSVC traffic statistics item at the SGSN side for fiveminutes. Calculate the value ofDownlink data kbytes sent to FR per NSVC x 8/(5 x
60) and compare the calculated value with the actual bandwidth. If the calculated
bandwidth does not exceed 70% of the actual bandwidth, the bandwidth is sufficient.
Otherwise, the bandwidth is not sufficient.
6. Packet loss causes the server to enter the congestion control state. Locate the node where
packet loss occurs. You can locate the node by capturing packets at each interface. IfTCP packets are lost in the transmission link or at an unacknowledged interface, for
example, the Gb interface, you must analyze packet loss by capturing packets through
the Ethereal. If a certain packet is lost at the MS side, you can locate the node where thepacket is lost by tracing the TCP packet sequence number, as shown in Figure 10:
2008-11-01 All rights reserved Page 27 of 38
-
7/27/2019 PS KPI
28/38
INTERNAL
Figure 10 Application layer data traced at the MS side
Figure 10 shows that the packet before packet 92064556 is lost. The size of each TCP packetis 1368 bytes, that is, 558 in hexadecimal. Thus, the sequence number of the lost packet is
92064556 558 = 92063FFE.
At the GGSN side, the Gi interface and Gn interface can be mapped to mirroring ports. Trace
the data of a single user on the SGSN. The traced data can be converted into an Etherealpacket capture file. Lost packets on these interfaces can be located through the TCP sequence
number.
For packet loss on the Gb interface, you can determine whether packet loss occurs based onthe messages traced on the Gb interface at the PCU side and whether the Nu is continuous. Nu
indicates the sequence number of an NS-PDU. Resolve the TCP packet header through thePCU. If one or multiple NS PDUs of the TCP packet are lost, the entire TCP packet is
discarded. The possible causes of packet loss on the Gb interface are as follows: interface
board, transmission link, frame check mode, and FR congestion control parameters if theintermediate network adopts the FR transmission mode. The frame check mode at the SGSN
side must be the same as that at the PCU side. As for the intermediate network exists between
the SGSN and the PCU, it is recommended that frame check be disabled. The congestion
control parameters of the FR include Bc, Be and CIR. In the period ofTc = Bc/CIR, if thedata transmission rate is higher than Bc but lower than Bc + Be, the packet may be discarded
during transmission; if the data transmission rate exceeds Bc + Be, the packet is certainly
discarded. In addition, even if the transmission rate does not exceed Bc, the transmission
network discards packets according to certain rules if the bandwidth of the FR transmissionnetwork is not insufficient.
The possible causes for the PCU to discard an LLC PDU are as follows:
The packet stays in the PCU buffer for more than 30 seconds. The PCU determines thatthe server will retransmit the packet due to RTO timeout if the packet is not sent after
such a period. This event is of extremely low probability.
An inter-PCU cell reselection occurs. This is called flush LL packet loss. The packet losscaused by the above two reasons can be reflected by the following parameters on the
PCU side: Number of Downlink LLC PDUs Discarded due to Timeout and
2008-11-01 All rights reserved Page 28 of 38
-
7/27/2019 PS KPI
29/38
INTERNAL
Number of Downlink LLC PDUs Discarded due to FLUSH (actually, it indicates thenumber of lost packets).
Packet loss due to the G-Abis interface is of low probability. If error bits exist on the G-Abis
interface and the BTS does not detect the error bits and sends the data over the Um interface,the data is lost as it fails to be assembled into an LLC PDU. The LLC layer usually usesunacknowledged mode, which may also lead to packet loss at the application layer.
Only unacknowledged ports lead to packet loss at the application layer. Even if packet loss occurs on the
Um interface or G-Abis interface, the RLC layer still sends the packet to the MS when the RLC uses theacknowledged mode.
Congestion control: The congestion control mechanism includes slow start and congestion prevention.The implementation method is that the TCP layer of the sending end maintains a congestion window and
slow start threshold. The congestion window is initially set with one data packet. In certain TCPapplications, this window is set with multiple packets, but the size cannot exceed 4380 bytes. The sum
of the congestion window threshold and maximum sequence number of an acknowledged data packet is
the upper limit of the sequence number of the packet sent through the TCP. When the congestionwindow is lower than the slow start threshold, the congestion window grows exponentially. When thecongestion window is greater than slow start threshold, the congestion window grows linearly each time
a packet is acknowledged. When packet loss occurs, the threshold of slow start is reduced by half. Whendata is re-sent due to timeout, the threshold of slow start is reduced by half, and the congestion window
is set with one data packet.
7. Weak transmission quality on the uplink leads to slow return of the TCP ACK message.
The server sends data only after receiving the ACK message. As a result, data
transmission is slow. The following situations may lead to the weak transmission qualityon the uplink:
Improper adjustment of the uplink coding scheme. Currently, the uplink coding
scheme is adjusted according to downlink coding scheme. If interference exists on the
uplink or the uplink level is weak, improper adjustment of the coding scheme maylead to failure of data transmission on the uplink. The uplink level can be checked
through the uplink and downlink balance traffic statistical item. The uplinkinterference can be checked through the Analyzed Measurement of InterferenceBand traffic statistical item. In the case that the uplink coding scheme is adjusted
according to downlink coding scheme, enter the super user mode, choose Configure
BSC Attributes > Software Parameters > Support EGPRS uplink MCS Dynamic
Adjust, and check whether the Support EGPRS uplink MCS Dynamic Adjust
parameter is set to Dl Ack with Downlink Signal Quality. If the value ofSupport
EGPRS uplink MCS Dynamic Adjust is the same as the downlink signal quality inthe DL ACK message, you can lower the uplink coding scheme by three classes by
entering the super user mode, choosing Configure BSC Attributes > Software
Parameter > DSP Control Table 2, and setting bit 5 to 1.
High RRBP frequency. Enter the super user mode and choose Configure BSCAttributes > Software Parameters > RRBP Frequency for EGPRS Downlink
TBF(Blocks). The default value ofRRBP Frequency for EGPRS downlink
TBF(Blocks) is 20. It is recommended that you set the parameter to a value larger
than 12. If the frequency is high, the MS cannot transmit uplink data blocks.
8. Improper settings of TCP parameters on the server or test PC. If the TCP window is set
to a small value, the TCP window easily gets congested. If the MSS is set to a small
value, the utilization rate is low. If the MSS is set to a large value, and the MTU on the
intermediate network is small, the IP packet is fragmented during transmission. Theparameter settings are described in section 4.2.
2008-11-01 All rights reserved Page 29 of 38
-
7/27/2019 PS KPI
30/38
INTERNAL
2.1.5 Abnormal Release of TBF
Abnormal release of TBF does not necessarily lead to packet loss, because the PCU saves the
data that is not sent or that is sent but not acknowledged by the MS in 30 seconds after the
abnormal release of TBF. The MS initiates a TBF reestablishment request shortly. The TLLIdoes not change at the time. The PCU can obtain the context of the MS according to the TLLI
and then sends the remainder to the MS.
How to determine whether a TBF is abnormally released? According to the TBF release flow,when a downlink TBF is normally released, the FAI in the Packet Downlink ACK/NACK
message is set to 1. When an uplink TBF is normally released, the FAIin the Packet Uplink
ACK/NACK message is set to 1. If the FAI is not set to 1, it indicates that the TBF is
abnormally released. If the network side sends a Packet TBF release message to release theTBF, you can infer that the TBF is abnormally released. If the cause value indicates that the
release is due to a normal event, you can infer that the abnormal release is caused by N3105
timeout.
Figure 11 Checking whether TBF is abnormally released through the ACK message
The impact of abnormal TBF release on the transmission rate is as follows: During the
release, data cannot be transmitted. The possible causes of abnormal TBF release are as
follows:
1. Timeout of N3101 and N3103
2. Timeout of N3105
3. Preemption of the control channel
4. Cell reselection
5. Internal processing faults
2.1.6 High Transmission Rate at the RLC Layer ButLow Transmission Rate at the Application Layer
This is usually caused by the incorrect operation of the tester. During the test, the tester mustdisable applications or services (for example, automatic update) that automatically initiate
2008-11-01 All rights reserved Page 30 of 38
-
7/27/2019 PS KPI
31/38
-
7/27/2019 PS KPI
32/38
INTERNAL
Figure 12 Comparison of the radio transmission environment before(upper) and after(lower) the network swapping
Before the network swapping, the cells with the C/I lower than 30 account for 60% of all
the cells. The cells with the C/I greater than 30 account for 40% of all the cells. After the
network swapping, the cells with the C/I lower than 30 account for 80% of all the cells,
and the cells with the C/I larger than 30 account for 20%. This indicates that the C/I fallsafter the network swapping. To deal with C/I decrease, you must check every cell along
the testing route to locate the cell where the C/I worsens, and then eliminate the
interference.
The fluctuation of the radio transmission environment is usually unpredictable, but
predictable in the scenarios such as the high-speed railway. In such scenarios, becausethe C/I changes fast, you must properly reduce the value ofBEP Period. In normal
scenarios, the recommended value is 5. In high-speed scenarios, the recommended value
is 4.
2. Cell reselection
Problems involved in cell reselection are as follows: disconnection delay caused by cellreselection, rate increase after cell reselection, service type detection during access to the
new cell after cell reselection, packet loss during cell reselection, and number of cell
reselections during the DT.
For disconnection delay caused by cell reselection, it is recommended that you enable
the NACC and Packet SI status functions to reduce the delay caused by cellreselection.
Rate increase after cell reselection. The application layer enters the congestioncontrol state due to disconnection after reselection. Congestion control slows down
data packet transmission. The low transmission rate in turn causes insufficient data at
the application layer. However, disconnection during reselection is inevitable. To
reduce the risks of this situation, a good transmission quality before reselection is
2008-11-01 All rights reserved Page 32 of 38
-
7/27/2019 PS KPI
33/38
INTERNAL
required. That is, the low data transmission rate before reselection and disconnectionduring reselection lead to severe congestion at the application layer. Therefore, the
wireless coverage and reselection parameters must be optimized to quicken the cell
reselection to a better radio transmission environment from a degrading radiotransmission environment.
Service type detection after reselection: This can be checked by the service typecontinuation function. The idea of this function is as follows: If an MS leaves the
original cell while performing the download service, the service type remains
unchanged after it enters the new cell. The MS is assigned with 4+1 channels in the
new cell.
Whether data loss occurs during reselection. Inter-PCU cell reselection may lead to
data loss. Therefore, for comparison DTs, try to avoid inter-PCU cell reselectionalong the DT route.
Number of reselections during DT. To minimize the number of cell reselection times,you should optimize the reselection parameters and neighboring cell configuration.
For the cells in which the MS stays a short time, it is recommended that you avoidarranging the testing route along such cells, especially those cells where ping-pong
reselections occur. Configure the neighboring cells that are missing. As shown inFigure 13, the neighboring relations and reselection parameters need to be optimized.
For details, see section 4.1.
Figure 13 Frequent reselections caused by improper neighboring cell configuration and
reselection parameter settings
2008-11-01 All rights reserved Page 33 of 38
-
7/27/2019 PS KPI
34/38
INTERNAL
3 Conclusion
3.1 Problem Identification
When rate problems are identified during test, you are advised to download large-sized files in
good radio transmission environment in idle hours to identify the problem. Then, compare the
current test method with others, find out their differences, and check if the differences can beoptimized.
1. Download large-sized files in places where the C/I is good and in idle hours. The test
result should be similar to that shown in Figure 4. Otherwise, you must identify the
problem.
The occupied channels are insufficient. It takes the MS about 4.5 seconds to occupied
four downlink channels.
Symptom: failure to assign more channels
Cause 1: Check whether the channels are sufficient by examining the
channel configuration.
Cause 2: Check the multislot capability of the MS. You can check the
multislot capability by checking the Packet resource request, 11-bit access
request, attach request, and the downlink data on the Gb interface when
Huawei SGSN is connected to Huawei PCU.
Cause 3: Loss of synchronization occurs on the channel. You can check
whether loss of synchronization occurs by checking the relevant alarms.
Cause 4: Abis interface resources or PCIC resources (in external PCU
mode) are insufficient.
Symptom 2: failure to occupy multiple channels steadily
Cause 1: The channels are pre-empted by voice services.
Cause 2: Loss of synchronization occurs on the channel. You can check
whether loss of synchronization occurs by checking the relevant alarms.
Low coding scheme
2008-11-01 All rights reserved Page 34 of 38
-
7/27/2019 PS KPI
35/38
INTERNAL
Symptom 1: long time occupation of a low coding scheme
Cause: Inadequate Abis interface timeslots
Solution 1: When the Flex Abis function is not enabled on the Abis
interface, set all the timeslots that have not been configured to idle. Solution 2: Increase the signaling link multiplexing degree to increase
the transmission rate on the Abis interface.
Solution 3: Enable the Flex Abis function.
Solution 4: Increase the transmission rate
Symptom 2: fluctuation of the coding scheme
Cause: Bit error on the G-Abis interface. You can check the bit error
through the G-Abis Interface Measurement traffic statistical item.
Cause 1: A transmission fault occurs. You can check for transmission
faults through local loop and remote loop at the TMU side.
Cause 2: A fault occurs on the interface board.
Block error
Cause 1: Bit error on the Um interface. You can check whether the C/I ratio on
the air interface is deteriorated.
Cause 2: Error frames and out-of-synchronization frames on the G-Abis
interface. You can identify this cause through the G-Abis Interface
Measurement traffic statistical item.
The MS is busy with other events. For instance, in the case that the number of
neighboring cells is large, and signals fluctuate greatly, if changes take place in
the six neighboring cells with the strongest signals, the MS decodes the systemmessages at the time when it receives data blocks. As a result, blocks are lost.
High proportion of control blocks
Cause 1: The MS subscription capacity is weak. You can check the MS PDP
context to identify the subscription capacity.
Cause 2: The acknowledged mode is used at the LLC layer. You can check the
MS PDP context to identify the mode.
Cause 3: The sending window stalls. This situation occurs only in GPRS
networks. You can identify this cause in the following ways: (1) Check whetherthe amount of data sent on the Gb interface is larger than that sent on the Uminterface during a certain period. (2) Increase the RRBP frequency and check
whether the situation is eased.
Cause 4: The performance of server is poor. You can check the server
performance by tracing the messages on the server.
Cause 5: The flow control performed on the Gb interface is improper. You can
trace the data on the Gb interface to check whether the flow control rate reported
by the PCU is higher than the sending rate on the Um interface.
Cause 6: Packet loss on the Gb interface and upper-layer NEs. You can use the
Ethereal to check packet loss at the MS side. Packet loss exists (packet loss
usually occurs on the unacknowledged interface) Then, capture the packets on
2008-11-01 All rights reserved Page 35 of 38
-
7/27/2019 PS KPI
36/38
INTERNAL
each interface to check on which interface packet loss occurs.
Abnormal release of TBF. By checking whether the value of the FAI cell in
Uplink ACK is 1, you can determine whether the uplink TBF is normallyreleased. By checking whether the value of the FAI cell in Downlink ACK is 1,
you can determine whether the downlink TBF is normally released. In addition,if the TBF release is caused by the PACKET TBF RELEASE message sent bythe network side, then the TBF release is abnormal. If the release cause cell
indicates normal release, the abnormal release of the TBF is caused by N3105
timeout.
The following are some possible causes of the abnormal release of TBF:
N3101 and N3103 timeout as the uplink wireless conditions worsen
Control channel being preempted
Windows failing to slide for several times
High rate at the RLC layer but low rate at the application layer
Cause: The MS performs other PS services during the test. You can use the
Ethereal to check for interaction data with other IP addresses, and identify theapplications or services that automatically initiate requests to access the Internet
according to the IP addresses. Then, disable these applications or services.
2. Compared with the CQT in idle hours, the CQT on downloading small-sized files inbusy hours may lead to lack of resources, including Abis interface resources, channel
resources, and PCIC resources for (in external PCU mode), and the channel may be
multiplexed by certain number of MSs. Compared with the CQT on downloading large-sized files, the CQT on downloading small-sized files may lead to lack of data at
application layer at the initial phase of data sending. This is because of the slow start
process during data sending at the TCP layer. In addition, timeslots are re-assigned afterthe service type is identified. This also lowers the transmission rate at the beginning ofdownload. Other problems are as follows:
Symptom 1: Compared with the test before the network swapping, the averagenumber of occupied channels is small. The causes of failure to assign channels are as
follows: no available channels, no Abis resources, and no PCICs (in external PCU
mode). Check the channel occupation during the test to check whether the channelsare sufficient, and check the traffic statistics and resource budget to see other
resources. If channels are insufficient, lower the threshold of dynamic channel
conversion.
Symptom 2: Compared with the test before the network swapping, the multiplexing
degree is high. In this case, you can lower the threshold of dynamic channelconversion and the channel multiplexing threshold.
3. For DTs in both idle and busy hours, even if the problems described in 1 and 2, the
adjustment of coding rate caused by the changes of radio transmission environment,disconnection for a period after cell reselection, and rate increase after cell selection are
all solved, the following problems may exist:
Symptom 1: The C/I ratio after the network swapping is worse than that before thenetwork swapping (based on the C/I WORST message exported from the TEMS). In
this case, network optimization is required.
Symptom 2:Cell reselection occurs frequently, especially in cells where the MS stays
a very short time. In this case, you must optimize the neighboring cells and
2008-11-01 All rights reserved Page 36 of 38
-
7/27/2019 PS KPI
37/38
INTERNAL
reselection parameters.
Symptom 3: Slow rate increase after cell reselection. This is due to the low rate
before cell reselection, and disconnection afterwards, which causes the server to enterthe congestion control state. In this case, you must optimize the reselection
parameters.
3.2 Routine Optimization
In routine optimization, you must consider factors that affect the transmission rate. The key
tasks in routine optimization are as follows:
1. Optimize the wireless conditions and neighboring cells in the same way as optimizationin the CS domain.
2. Ensure and assign resources. Identify the bottleneck of resources. The bottleneck must bethe most precious resources, which in most cases, are wireless resources. In this case,
you must guarantee transmission resources on interfaces, such as the Gb interface. In
addition, CS and PS services share the wireless and Abis resources now. You mustoptimize assignment of these resources and the management parameters of these
resources to maximize the bandwidth for PS services while ensuring the bandwidth for
CS services.
3. Settle the problems that are common on the entire network, such as high bit error rate at
the Gabis interface, and packet loss at the GB interface. Section 4.1describes the
common parameters that affect download rate.
For the swapped network, you must obtain the following information:
Network structure. For example, whether DXX devices are used on the Abis interface for
timeslot cross and whether the Gb interface uses the FR transmission mode.
Resource allocation, including the number of timeslots assigned to the Abis interface for
each BTS by the original network, relevant resource management parameters, number of
timeslots assigned to the Gb interface, and FR congestion control parameters when theGb interface uses the FR transmission.
Traffic statistics on the Gb interface and Um interface of each PCU on the originalnetwork. The statistics are used for traffic comparison.
Test log.
After the network swapping, make routine observation on the following traffic statistics and
alarms:
Check whether an alarm related to out-of-synchronization channel is generated.
Check whether an alarm related to out-of-synchronization with the remote end is
generated on the Gb interface.
Check the G-Abis Interface Measurement, TBF establishment success rate, calldrop rate, and PS Channel Measurement.
2008-11-01 All rights reserved Page 37 of 38
-
7/27/2019 PS KPI
38/38
INTERNAL
4 Appendices
4.1 Appendix 1: Parameter Description
Setti ngs ofDownload-Related
4.2 Appendix 2: Optimization Manual forParameters on Server and Test PC
GPRSEGPRS TestFTP Parameter Opti