LTE-Optimization

Slide titleIn CAPITALS

44 pt

Slide subtitle 20 pt

LTE Optimization

Slide title 30 pt

Text 18 pt

Bullets level 2-516 pt

›!"# $%&'()*+,-./0123456789:;<=>?@ABCDEFGHIJKLMNOPQRSTUVWXYZ[\]^_`abcdefghijklmnopqrstuvwxyz{|}~¡¢£¤¥¦§¨©ª«¬®¯°±²³´¶·¸¹º»¼½ÀÁÂÃÄÅÆÇÈËÌÍÎÏÐÑÒÓÔÕÖ×ØÙÚÛÜÝÞßàáâãäåæçèéêëìíîïðñòóôõö÷øùúûüýþÿĀāĂăąĆćĊċČĎďĐđĒĖėĘęĚěĞğĠġĢģĪīĮįİıĶķĹĺĻļĽľŁłŃńŅņŇňŌŐőŒœŔŕŖŗŘřŚśŞşŠšŢţŤťŪūŮůŰűŲųŴŵŶŷŸŹźŻżŽžƒȘșˆˇ˘˙˚˛˜˝ẀẁẃẄẅỲỳ–—‘’‚“”„†‡•…‰‹›⁄€™−≤≥fifl

Why learn about Throughput Troubleshooting

› LTE provides data, lots of data

› Throughput is shared in time and frequency

› Users notice throughput problems

› Learn to troubleshoot the LTE RAN for throughput problems

› Learn to isolate the domain causing throughput degradation


44 pt


> Overview

Slide title 30 pt

Text 18 pt



Agenda

1. Overview

2. Initial Checks

3. Radio Analysis

4. Transport Analysis

5. E2E Analysis


44 pt


LTE RBS User plane Overview


44 pt


Initial Checks

Slide title 30 pt

Text 18 pt



Initial checks

› For throughput issues, some essential checks are required:

1. Network changes & Basic troubleshooting

2. PC/Server settings

3. UE categories

4. UE subscriber profile

5. RBS parameters

6. Enabled features

Slide title 30 pt

Text 18 pt



NW Changes and Basic Troubleshooting

› Network/node changes can affect network throughput› Some common examples include:– Network configuration changes (e.g. adding/changing/removing hardware)– RBS parameter changes (all MOs under ENodeBFunction, system

constants, EricssonOnly hidden parameters, e.g. DataRadioBearer)– IP address plan changes– Transport Network changes (add/reduce capacity on TN)– DNS updates– Hint:› To see RBS level changes (MOs/parameters): Moshell> lgo› To capture detailed RBS level logs: Moshell> dcg

› Basic troubleshooting checks include:– Alarm, event and system log checks– MO health status

Slide title 30 pt

Text 18 pt



PC/Server Settings

› Determine the applications or tools used in testing/monitoring throughput

› Confirm the end user PC settings:– Laptop specification can impact throughput (processors, memory, USB bus, HDD

speed, plugged into AC power, etc)– MTU settings in PC (1360 optimal for eNB in L11A to prevent fragmentation)– Throughput monitors (e.g. Netpersec, only good for downlink UDP measurements,

uplink must be measured at receiving side for UDP)– TCP enhancements in Vista (experimental), Vista should “auto-tune”.

› Confirm server settings:– FTP server configuration– Linux TCP setting/guide– iperf (UDP & TCP) – be sure to use packet size 1360 for UDP (not default 1470).– Always check first with UDP rather than TCP, as UDP is less prone to display

problems as a result of jitter variations and packet loss.

Slide title 30 pt

Text 18 pt



UE Categories

UE Category Maximum number of DL-SCH transport block bits received within a TTI

Maximum number of bits of a DL-SCH transport

block received within a TTI

Total number of soft channel

bits

Maximum number of supported layers for

spatial multiplexing in DL

Category 1 10296 10296 250368 1

Category 2 51024 51024 1237248 2

Category 3 102048 75376 1237248 2

Category 4 150752 75376 1827072 2

Category 5 299552 149776 3667200 4

› The UE Category limits throughput possibilities

› 5 UE Categories are defined in 3GPP TS 36.306

› The UE-Cat is sent in the UE Capability Transfer procedure (RRC UECapabilityInformation)

› The COLI ue command provides detailed capability info (KO) for connected UEs

UE Category Maximum number of bits of an UL-SCH

transport block transmitted within a

TTI

Support for 64QAM in UL

Category 1 5160 No

Category 2 25456 No

Category 3 51024 No

Category 4 51024 No

Category 5 75376 Yes

DL UL

Slide title 30 pt

Text 18 pt



UE Subscriber profile

› End User (EPS User) subscription data is stored in the HSS

› The EPS User Profile data is identified by its IMSI number

› The profile consists of:– MSISDN number– Operator Determined Barring (ODB)– APN Operator Identifier Replacement– Subscribed Charging Characteristics– Aggregate Maximum Bit Rate (AMBR)

› Max requested bandwidth in Downlink› Max requested bandwidth in Uplink

– RAT frequency selection priority– APN configuration profile:

› Default Context Identifier (default APN for the EPS User)› APN Configuration (every APN associated to the EPS User)

Slide title 30 pt

Text 18 pt



RBS PARAMTERS TN

› TN MO parameters:– GigabitEthernet=1

› actualSpeedDuplex – if you see half-duplex, it could be a problem with auto-negotiation

› dscpPbitMap (QoS mapping from L3 to L2)– IpInterface=2 (rec. MO id for Signalling and Payload)

› vLan/vid (true/false and vlan id)– IpAccessHostEt=1

› ipAddress (X2/S1 control/user plane termination)– IpSyncRef (if NTP synchronisation is used)

› syncStatus should be OK– Synchronization=1

› nodeSystemClock should be in LOCKED_MODE.› syncReference should show the correct reference (NTP or GPS) active and

configured

Slide title 30 pt

Text 18 pt



Enabled Features

› User throughput can be limited by the available/installed licenses

› The following features directly impact end user throughput– Downlink/Uplink Baseband Capacity– Channel Bandwidth (5, 10, 15 and 20) MHz– 64-QAM DL / 16-QAM UL– Dual Antenna DL Performance Package

› To quickly check active licenses (including states):– moshell> inv

Licensing (9/1551-LZA 701 6004 )

Slide title 30 pt

Text 18 pt



Expected Throughput (Simplified)

dlCyclicPrefix = 15 KHz => 7 OFDM symbols

Resource Elements (RE) per Resource Block(7 OFDM symbols x 12 SubCarriers)RE per SB2 x RBRS RE (per RB)RS RE (per SB)

Control Region Size (CRS) in OFDM symbolsnrOfSymbolsPdcch 1 2 3 1 2 3RE per CRS(OFDM*12 - 4 RS Tx)(OFDM*12 - 8 RS MIMO) 8 20 32 16 40 64Tot Num RE per SB available for PDSCH(best case w/o SCH/BCH) 144 132 120 288 264 240Bits per SB - QPSK (2) 288 264 240 576 528 480Bits per SB - 16QAM (4) 576 528 480 1152 1056 960Bits per SB - 64QAM (6) 864 792 720 1728 1584 1440

Max Theoretical L1 Thrpt (Mbps)

20 MHz => 100 RB (64 QAM) 86.4 79.2 72 172.8 158.4 144

15 MHz => 75 RB (64 QAM) 64.8 59.4 54 129.6 118.8 108

10 MHz => 50 RB (64 QAM) 43.2 39.6 36 86.4 79.2 72

5 MHz => 25 RB (64 QAM) 21.6 19.8 18 43.2 39.6 36

Tot Num RE per SB available for PDSCH(worst case with SCH/BCH in SB)SCH = 24, BCH = 4 x 12 - 4 per CW 76 64 52 152 128 104Bits per SB - (QPSK) 152 128 104 304 256 208Bits per SB - (16QAM) 304 256 208 608 512 416Bits per SB - (64QAM) 456 384 312 912 768 624

Tx Diversity 2x2 MIMO

84

DL Scheduling Block (SB) -> Bit calculation(Normal CyclicPrefix)

168

16 32

168 3368 16

Slide title 30 pt

Text 18 pt



Identify the domain

› Further analysis required:– Our basic checks have come up short– Throughput issues exist that require advanced/additional analysis

› Analysis steps to perform:– Single UE call scenario– Send UDP type traffic in DL/UL direction (e.g. Iperf)– Monitor close to or on the RBS (e.g. Wireshark)– Optionally use a radio monitor (e.g. TEMS)

› Decide - Radio or Transport analysis:– Radio issues provide more control for LTE RAN analysis– Transport issues blend/carry-on towards core elements


44 pt


Radio Analysis

Slide title 30 pt

Text 18 pt



Radio Analysis

› From the domain analysis previously, we believe the Radio may be affecting user throughput

› We’ve previously ruled out configuration and MO status using the basic checks

› The following slides will cover the various components which make up the radio domain and help to pinpoint the source of poor throughput.

› We rely on the baseband scheduler traces and signal traces (mtd) between blocks.

Slide title 30 pt

Text 18 pt



Radio Analysis

› Ericsson’s LTE Baseband provides a detailed mechanism for tracing the complete L1 and L2 interaction, including MAC scheduling decisions and L1 decoding results.

› Using this information we can further isolate the cause of the problem and pinpoint either:– UE problem

› Cannot detect ACK/NACK?› Invalid UE reports?

– Uu air interface problem– eNB problem

› Incorrect setting or non-optimal combination of settings› Scheduling abnormality› Limitation in current eNB software

– eNB northbound problem› S1 user plane› Application Server› Core network, SASN/SGW, etc

Slide title 30 pt

Text 18 pt



RADIO ANALYSIS

› To perform targeted radio analysis, it’s useful to know radio aspects specific to the following traffic scenarios:

1. Downlink

2. Uplink

3. Both uplink and downlink

› Post-processing tools will be briefly demonstrated

Slide title 30 pt

Text 18 pt



Radio Analysis - Downlink

› Areas of analysis for Downlink:– CQI (Channel Quality Index) and RI (Rank Indicator) reported from UE.– Transmission Mode: MIMO (tm3) vs. TxD (tm2) vs. SIMO (tm1)– MCS vs. number of assigned PRBs vs. assignable bits in scheduler– UE Scheduling percentage of TTIs (how often is the UE scheduled)– CFI (number of OFDM symbols for PDCCH) vs. MCS vs. % scheduling– HARQ– RLC retransmissions

Slide title 30 pt

Text 18 pt



Radio Analysis – Uplink

› Areas of analysis for Uplink:– Uplink scheduling overview– BSR (Buffer Status Report)– PHR (Power Headroom Report) – is the UE at maximum power?– Cell bandwidth vs. maximum allowable PRBs– Link Adaptation– MCS available and 16QAM– PDCCH SIB scheduling colliding with UL grant– HARQ (less important, because we can measure SINR)

Slide title 30 pt

Text 18 pt



Radio Analysis DL – CQI/RI and TM

› The eNB needs knowledge of the SINR conditions of downlink transmission to a UE in order to select the most efficient MCS/PRB combination for a selected UE at any point in time.

› Channel Quality Index (CQI):– Is a feedback mechanism from UE to eNB– Informs eNB of current channel conditions as seen at UE– Directly maps to 3GPP defined modulation/code rate (TS36.213 Table 7.2.3-1)

› Defined as the highest coding rate the UE could decode at 10% BLER on HARQ rv=0 transmission

– CQI 1-6 map to QPSK– CQI 7-9 map to 16QAM– CQI 10-15 map to 64QAM

› Rank Indicator (RI)– Is a feedback mechanism from UE to eNB– Informs eNB whether UE can successfully decode RS from 1 or 2 (or more)

antennas.– eNB scheduler uses this feedback to transmit with either:

Slide title 30 pt

Text 18 pt



CQIpolling

Radio analysis DL – CQI/RI and TM› The UE measures DL channel quality

and reports to eNodeB in the form of Channel Quality Information (CQI)

› The average CQI (periodic-CQI reporting) for the whole band (wide-band CQI) is reported periodically on PUCCH (or on PUSCH if user data is scheduled in that TTI) with configured periodicity.

› Sub-band CQI (aperiodic-CQI reporting) is reported when requested by the eNB. This report is for the PDSCH. Report sent on PUSCH.– CQI polling is triggered on demand

by eNB based on DL traffic activity.

› When 2 antennas are configured, Rank Indicator is also reported. Precoding Matrix Indicator (PMI) also reported in case of transmission mode 4 (not in L11A).

CQI

DL

fre

qu

en

cy

ba

nd

PUCCH

PUCCH

PU

SC

H

Slide title 30 pt

Text 18 pt




› In order to transmit with MIMO (OLSM) we should check the following:– eNB cell is configured with two working transmit antennas.

› Check EUtranCellFDD::noOfUsedTxAntennas > 1› L11A GA (default) system constant SC125:3 means that tm3 is used in case 2

TX antennas are defined.› If only one TX antenna is configured, then tm1 is used› In order to force Transmit Diversity (i.e. prevent OLSM), SC125:2 must be set

– UE CQI/RI report from UE shows RI > 1› Rank 1: TxDiversity (transmission mode 2, tm2)› Rank 2: MIMO (Open Loop Spatial Multiplexing in L11A) (transmission mode 3,

tm3)

› mtd peek -ta ulMacPeBl -signal LPP_UP_ULMACPE_CI_UL_L1_MEAS2_DL_IND -dir OUTGOING

– This signal (from L1 to MAC scheduler) shows the reported CQI and RI– (also shows HARQ ACK/NACK for downlink data transmission)– (also shows rxPowerReport and timingAdvanceError)

Slide title 30 pt

Text 18 pt



Radio analysis DL – CQI/RI and TM

› cfrFormat = 4 consists of:– 4 bit Wideband CQI (i.e. CQI across whole bandwidth)– Up to 13 subband CQI differentials (depends on bandwidth of cell)

› Subband CQI (3GPP TS36.211 Ch 7.2.1)– RBG width depends on bandwidth:› 3 & 5MHz – subband width 4 PRBs› 10MHz – subband width 6 PRBs› 15 & 20MHz – subband width 8 PRBs

– Subband Differential mapping, see table below:

Slide title 30 pt

Text 18 pt




› RBG for Resource Allocation Type 0– Defined in 3GPP TS36.213 Ch 7.1.6.1– One bit used to represent a certain number of consecutive PRBs– 1.4MHz is RBG size 1– 3 & 5MHZ is RBG size 2– 10MHz is RBG size 3– 15 & 20MHz is RBG size 4

Slide title 30 pt

Text 18 pt



Radio Analysis DL – Assignable Bits

› If UE is sending with high CQI (in the range 10-15) and RI=2 but throughput is still very low, then the next check should be assignable bits.

› Assignable bits means the amount of data in the downlink buffer available for the scheduler to schedule for this UE.

› A classic symptom of low assignable bits is that the UE is scheduled with a high MCS but a low number of PRBs.– The scheduler always attempts to send with the highest possible MCS and

least number of PRBs so that left-over PRBs could be assigned to another UE.

› Another symptom is that the UE is not scheduled every TTI (and nothing else is available to schedule).

Slide title 30 pt

Text 18 pt



Radio Analysis DL – Assignable Bits

› Possible causes for low assignable bits:1. RLC STATUS messages are not being received fast enough and RLC

buffers are full. › Until RLC STATUS ACK messages are received, already transmitted

RLC SDUs are kept in memory in UE and/or eNB› Check for RLC DISCARDs but low (or 0) assignable bits

2. Data received from core network is not enough to fill the RLC buffers in eNB. › Check that non-TCP based traffic is not being sent with too large

packet size. For iperf based traffic, recommended size 1360 bytes (default is 1470).

› Set MTU of 1360 in UE (or UE laptop).› RLC DISCARDs will trigger TCP congestion control and lower thpt.

› In L11A the default RLC buffer size per RB is 750 IP packets– Trace discards with lhsh gcpu00768 te e all UpDlPdcpPeFt_DISCARD – Discards on UDP traffic will not affect throughput– Discards on TCP traffic will trigger TCP congestion control (lower thpt.)

Slide title 30 pt

Text 18 pt



Radio Analysis DL – CFI and Scheduling

› SIBs require PDCCH resources› Typically SIBs consume 4 or 8 CCEs of PDCCH resources.› If a UE is in good SINR conditions, the scheduler may allocate only

one CCE for that UE. – In that case, because of limited positions in PDCCH, it is quite likely that a

PDCCH collision occurs (especially in low system bandwidths)

› If a UE is in bad SINR conditions, the scheduler may allocate a large number of CCEs for that UE (2 or 4 or 8 CCEs)– Depending on the configured CFI there may only be common search space

available or it may still collide with other PDCCH users.

› See Radio Analysis UL – PDCCH slides for more details

Slide title 30 pt

Text 18 pt



Radio Analysis Dl – HARQ

› Each transport block transmission is represented as a HARQ process.– Each HARQ process data is held in memory until NDI is toggled (i.e. New data is to

be sent). – This allows fast retransmission of erronerously received data.

› The schedulers representation of an HARQ process is as follows:– Feedback status

› (ACK, NAK, DTX, PENDING)– TBS – transport block size– MCS – modulation and coding scheme– RV – redundancy version. HARQ has 4 redundancy versions, rv0, rv2, rv3, rv1.– NDI – New Data Indicator (physical layer bit toggled for new data).

› Do not confuse with newDataFlag which is scheduler internal flag where 1 means new data and 0 means retransmission.

– Number of transmission attempts (max 4 transmissions in L11A default paramters)

› In case of rank 2 spatial multiplexing there are 16 HARQ process per UE instead of 8, but there are two processes that share the same ID– Scheduler sees them as separate processes that are coupled to each other

Slide title 30 pt

Text 18 pt



Radio Analysis Dl – HARQ Example

› The following slides will show an example of tracing out downlink HARQ– Initial downlink grant is sent with rv=0 (MIMO, 2 codewords)

› SFN 280/subframe 8

– HARQ NACK received on both code words› SFN 281/subframe 2 (DL Grant + 4TTI)

– First retransmission sent with rv=2› SFN 281/subframe 6 (8 TTI past initial transmission is earliest occasion)

– HARQ ACK received on both code words› SFN 282/subframe 0 (DL Grant ReTx + 4TTI)

Slide title 30 pt

Text 18 pt



Radio Analysis DL – RLC

› RLC retransmissions are triggered:1. When HARQ fails to transmit a transport block within the maximum number of

configured retransmissions– Default number of HARQ transmissions is 4 in L11A

2. If RLC STATUS messages are not received within the time frames configured

› RLC STATUS messages are sent between peer nodes (eNB and UE) to inform about lost RLC packets. They can be traced out using– mtd peek -ta dlRlcPeBl -si UP_DLRLCPE_FI_STATUS_FOR_DL_TRAFFIC_IND

› Check:– ACK_SN should be increasing, otherwise RLC buffers are not released– NACK_SN indicates RLC retransmissions (occasionally is OK)– DataRadioBearer::tStatusProhibit governs how often RLC STATUS messages may

be generated, default is 25ms in L11A. › A too low value will produce too many RLC control messages› A too high value may cause RLC buffers to become exhausted

Slide title 30 pt

Text 18 pt



Radio Analysis – Uplink

› Areas of analysis for Uplink:– Uplink scheduling overview– BSR (Buffer Status Report)– PHR (Power Headroom Report) – is the UE at maximum power?– Cell bandwidth vs. maximum allowable PRBs– Link Adaptation– MCS available and 16QAM– PDCCH SIB scheduling colliding with UL grant– HARQ (less important, because we can measure SINR)

Slide title 30 pt

Text 18 pt



Radio Analysis UL – UPlink Scheduling

UL› Scheduling request, SR (PUCCH) UE requests UL resources

UL grant

eNodeB

UL scheduler

Ue

Scheduling Request (SR)

Buffer status report (BSR)

Data

Channel sounding

Channel state info

› Data is transmitted (PUSCH)

› UL Grant (PDCCH)Scheduler assigns initial resources

› Buffer status report (PUSCH) transmitted in UL

› UL grant (PDCCH) transmitted (valid per UE)

UL grant

› Channel sounding

Slide title 30 pt

Text 18 pt



Radio Analysis UL – PUCCH and PUSCH

› PUCCH takes a minimum 1 PRB on each side of the uplink band for uplink control signalling, reducing the size of PUSCH– E.g. 5MHz bandwidth, 25 PRBs available. Minimum 2 PRBs for PUCCH.– 23 PRBs available for PUSCH

0 1 2 3 4 5 6 7 8 9Time (ms)

Radio Frame

Cell Bandwidth PUSCH – Used for UE data

scheduling and UL RA msgs

PUCCH – Semi-static allocation of CQI, SR, ACK/NAK

PUCCH – Semi-static allocation of CQI, SR, ACK/NAK

PUCCH

PUCCH

PUSCH

Slide title 30 pt

Text 18 pt



Radio Analysis UL – Link Adaptation

Goal: Select MCS for a certain allocation size to maintain the target BLER (10%) for the first transmission

Inputs to Uplink Link Adaptation are:

UL interference power:

LPP_UP_ULCELLPE_CI_CELL_STATUS_REPORT_IND outgoing from ulCellCeBl

Received power of UE (across traffical PUSCH PRBs):

LPP_UP_ULMACPE_CI_UL_L1_MEAS2_UL_IND outgoing from ulMacPeBl

PHR reports & HARQ CRC (BLER):

LPP_UP_ULMACPE_CI_UL_MAC_CTRL_INFO_IND outgoing from ulMacPeBl

Inpu

t to

Lin

k A

dapt

atio

n

ulL1Meas2UlInd(rxPower(-140 .. 0 dB)

ulMacPe

run every subframe UE transmitts

cellStatusReportInd(interferencePower (-125 .. -80dB)

ulCellCe

run every subframe

ulL1MacCtrlInfoInd(powerHeadroom(0 .. 63dB))

ulMacPe

run every periodicPhrReport

ulL1MacCtrlInfoInd(CRC)

ulMacPe

run every subframe UE transmitts

Slide title 30 pt

Text 18 pt



Radio Analysis UL – Link Adaptation

› Check for:– High values of UL interference› Could there be some external interferer?› Are the values of pZeroNominalPusch in neighbour cells too high?

– rxPower too low› Target is EUtranCellFDD::pZeroNominalPusch. Is it set too high?

– PHR shows UE at maximum Tx power› Is EUtranCellFDD::pZeroNominalPusch too high causing UE to exceed

maximum transmit power?› Closed-loop power control TPC ignored by UE?

– Low values of SINR› Is EUtranCellFDD::pZeroNominalPusch too low?› Closed-loop power control TPC ignored by UE?

Slide title 30 pt

Text 18 pt



Radio Analysis UL – PDCCH

0 1 2 3 4 5 6 7 8 9Time (ms)

Radio Frame

PDCCH carries both the UL (PUSCH) assignment and DL (PDSCH) assignment.

In case many PDCCH CCEs are used for DL transmission (e.g. SIB with 8 CCEs) it may be that UL grant is not possible to be scheduled in this TTI for a single UE!

PUSCH

UL subframe (4 TTI later)

DL subframe (current)

PDSCH

PD

CC

H

Note: PCFICH and PHICH multiplex into the DL subframe red area marked PDCCH

Slide title 30 pt

Text 18 pt



Radio Analysis UL – PDCCH

From KO link: PDCCH visualisation KO

Search space for 1 CCE completely overlaps 8 CCE search space.

In this example, DL SIB transmission completely prevents any UL grant for this UE RNTI 516 in subframe 5

Slide title 30 pt

Text 18 pt



Radio Analysis UL – HARQ

› LTE defines uplink with synchronous HARQ to reduce PDCCH signaling load and simplify the uplink HARQ processing

› Example 1, successfully received PUSCH data:– Subframe n: UL grant sent to UE– Subframe n+4: PUSCH data received (rv=0)– Subframe n+8: ACK sent, UL grant with New Data Indicator toggled – Subframe n+12: new PUSCH data received (new HARQ process)

› Example 2, HARQ retx:– Subframe n: UL grant sent to UE– Subframe n+4: PUSCH data received (rv=0)– Subframe n+8: NACK sent, NO UL grant is signaled on PDCCH– Subframe n+12: PUSCH data received (rv=2) (same HARQ process)– Subframe n+16: ACK/NACK, etc up to max number of retx

Slide title 30 pt

Text 18 pt



Radio Analysis – Post-Processing Tools

› 3GPP has specified L1 messages in order to reduce the bits required for transmission on the air interface.– These formats can be difficult to read

› For this reason, many values in the traces are presented in formats which require conversion to human readable formats, for example:– PRBs allocated in DL/UL grant messages– PHR values – BSR values– SINR– MIMO HARQ feedback, etc..

› Tools exist to perform these conversions and compact the data presentation to the end user– One such tool is bbfilter or scheduling_filter.pl– Check the flowfox web page for details

Slide title 30 pt

Text 18 pt



Radio Analysis - Summary

› Areas of analysis for Downlink:– CQI / RI (Rank Indicator) reported from UE.– Transmission Mode (MIMO, TxD, SIMO)– MCS vs. number of assigned PRBs vs. assignable bits in scheduler– UE Scheduling percentage of TTIs (how often is the UE scheduled)– PDCCH CFI and scheduling impacts– HARQ– RLC retransmissions

› Areas of analysis for Uplink:– BSR (Buffer Status Report)– PHR (Power Headroom Report) – is the UE at maximum power?– Cell bandwidth vs. maximum allowable PRBs– Link Adaptation– MCS available and 16QAM– PDCCH SIB scheduling colliding with UL grant– HARQ (less important, because we can measure SINR)


44 pt


Transport Analysis

Slide title 30 pt

Text 18 pt



Transport Analysis

› End user data is transferred over the S1-U interface

› Several Transport Network topologies (L2/L3) provide great flexibility in design

› Several router redundancy methods are supported

› Transport network dimensioning provides insights into the peak provisioning on the S1 link

› The LTE RBS is a QoS enabler, providing end user and transport network QoS differentiation

› Performance management counters and tracing provide us with powerful node observability methods

Slide title 30 pt

Text 18 pt



Transport Topology

Slide title 30 pt

Text 18 pt



Transport Topology

› No strict requirements on using a L2 switched or L3 routed LTE RAN transport network

› No specified topology requirement

› A router is required in the network, but LTE RAN transport network does not have to be L3

› Network design is important (number of hops for L3 vs. size of broadcast domain for L2)

› This topology flexibility could complicate troubleshooting efforts depending on the nodes involved (say 3PP support is required)

Slide title 30 pt

Text 18 pt



Router Path Supervision (RPS)

› RPS provides router redundancy for S1/X2 traffic

› RPS is configurable via the IpInterface MO

Slide title 30 pt

Text 18 pt



Transport Dimensioning

› Dimensioning of the northbound transport network will impact achievable end user throughput rate

› LTE RBS transport network dimensioning process (mobile backhaul):

› Dimensioning is based on payload only!

Determine bandwidth needed for last mile

Determine cell thpt in a loaded network and avg. cell thpt during busy hour

Calculate agg. bandwidth required in mobile backhaul

Slide title 30 pt

Text 18 pt



Dimensioning methods

Method Description

Overbooking Allows more users than the dimensioned quantity, using the rationale that only a subset of users is allocating bandwidth at the same time.

Overdimensioning Calculates the dimensioned bandwidth required by multiplying the average requirement by an overdimensioning factor.

Peak allocation Uses the maximum throughput capacity as the dimensioned link capacity. The link is dimensioned for the maximum possible bit rate.

Overprovisioning Monitors the link use. When a predefined use limit is reached on the link, a capacity upgrade is initiated. A general rule is that the use limit is set to 50%.

Slide title 30 pt

Text 18 pt



Transport Aggregation

Input (assumptions) 20 MHz Cell

Cell Peak Rate 150 Mbps

Cell Throughput in a Loaded Network

35 Mbps

Peak load for 3x1 in a Loaded Network

~100 Mbps

S-GW/PDN GW

A3

A1

A1

S-GW/PDN GW

RBS

RBS

RBS R

BS

RBS

RBS

RBS

RBS

RBS

A1

A2 A2

Dimension for: ΣA2 × 0.8

BH displacementfactor

Dimension for peak rate to 1 cell= 150 Mbit/s

Dimension for ‘eNodeB throughput in a loaded network for a 3x1 configuration’ = 100 Mbit/s per eNB

Dimension for ‘Average eNodeB throughput during Busy Hour’ = 50 Mbit/s per eNB

Slide title 30 pt

Text 18 pt



Quality of service (qos)

› Transport network QoS is a part of the complete LTE QoS concept

› QoS, in an IP transport network, is the ability to treat packets/frames differently based on their content

› Without QoS, each packet/frame is given equal access to the network resources

Slide title 30 pt

Text 18 pt



QOS for user plane Bearers

TransportRANTerminal

Gateway(Bearer Policy

Enforcer)

Service 1(e.g. Internet)

Service 2(e.g. P2P File Sharing)

Service 3(e.g. IMS-Voice or MTV)

Default Bearer (QoS via MME)

Dedicated Bearer (QoS via PCRF)

Service Data Flow (SDF)

IP Address

Further Reading:3GPP TS 23.401

Slide title 30 pt

Text 18 pt



Quality Class Indicators

› A Quality Class Indicator (QCI) is used to signal the QoS requirements of a bearer› 23.401 defines 9 standard QCIs, each one with specific characteristics› Operators may define proprietary QCIs to introduce new services

QCI Resource Type Priority Packet Delay BudgetPacket Loss

RateExample Services

1

GBR

2 100 ms 10-2 Conversational Voice

2 4 150 ms 10-3 Conversational Video (Live Streaming)

3 3 50 ms 10-3 Real Time Gaming

4 5 300 ms 10-6 Non-Conversational Video (Buffered Streaming)

5

Non-GBR

1 100 ms 10-6 IMS Signaling

6 6 300 ms 10-6

- Video (Buffered Streaming) - TCP-based (e.g., www, e-mail, chat, ftp, p2p file sharing, progressive video, etc.)

7 7 100 ms 10-3

- Voice, - Video (Live Streaming) - Interactive Gaming

8 8

300 ms 10-6

- Video (Buffered Streaming) - TCP-based (e.g., www, e-mail, chat, ftp, p2p file sharing, progressive video, etc.)9 9

Slide title 30 pt

Text 18 pt



Transport Network QoS – Layer 3

Site Infrastructure, IP-backbone and RAN Transport EPC

Application

IP

PPPoE Ethernet

Ethernet

IP and Ethernet

IP Transport Network

IPx is Interface between node and IP transport network

Version4

Header Length4

DiffServ7

Total Length16 20 bytes

Further Reading:RFC 2474

Differentiated Services Code Point (DSCP)

› QCIs are mapped to IP layer Differentiated Services Code Points (DSCP)

Slide title 30 pt

Text 18 pt



Transport Network QoS – Layer 2

Site Infrastructure, IP-backbone and RAN Transport EPC

Application

IP

PPPoE Ethernet

Ethernet

IP and Ethernet

IP Transport Network

IPx is Interface between node and IP transport network

Preamble7

DA6

SA6

TPI2

TAG2

Type2

Data46 to 1500

CRC4

SFD1

User Priority3bits

CFI1bit

VLAN ID 12bits

› Use of P-Bits allows prioritizing different types of traffic› Priority queuing, enabling some Ethernet frames to be forwarded ahead of others within a

switched Ethernet network› Frames can be assigned to different scheduler queues› Maintain the same value throughout the IP Network to deliver the same QoS

Further Reading:IEEE 802.1p

Priority Bit (P-bit)

Slide title 30 pt

Text 18 pt



QOS Mapping

› LTE RAN Quality of Service (QoS) ensures the LTE bearer and transport level service requirements

EthernetFrame

IP Packet

QoS Profile

DATA

DATA

QCI

AMBR ARP

IP Header

Ethernet Header

DSCP

Mapping:Takes place in RBS / EPC

Mapping:Edge devices handling L2/L3 payload

P-bit

Slide title 30 pt

Text 18 pt



LTE RAN QoS Configuration

› LTE RAN complies with 3GPP TS 23.203 and IEEE 802.1p:

› Using these MOs we are able to map out the quality of service properties defined in the RBS

===============================================================

MO dscp priority qci

===============================================================

QciTable=default,QciProfilePredefined=qci1 46 2 1





QciTable=default,QciProfilePredefined=default 0 10 0





===============================================================

====================================================

MO Attribute Value

====================================================

Subrack=1,Slot=1,PlugInUnit=1,ExchangeTerminalIp=1,GigaBitEthernet=1 dscpPbitMap t[64] =

>>> Struct[0] has 2 members:

>>> 1.dscp = 0

>>> 2.pbit = 0

... truncated ...


>>> 1.dscp = 46

>>> 2.pbit = 6


>>> 1.dscp = 47

>>> 2.pbit = 0

Transport Network Configuration (39/1553-HSC 105 50/1)

Slide title 30 pt

Text 18 pt



GTP-U - User Plane

› GTP-U is used as the User Plane protocol over the S1 and X2 interfaces (defined in 29.060)

› GTP-U carries the end user data by forming tunnels towards the core network S-GW (transports user IP payload)

› IP fragmentation is to be avoided (MTU size should be set correctly)

› Configuration aspects on the RBS are minimal (no MO models this layer)

GTP-U

UDP

IP

Data Link

Physical

GTP-U

UDP

IP

Data Link

Physical

S1-U

teid, ip address, portteid, ip address, port

Slide title 30 pt

Text 18 pt



Transport Network performance

› Transport Network performance visibility is available via the GigaBitEthernet MO and IpAccessHostEt pm counters– Moshell> pmom GigabitEthernet|IpaccessHostEt

› The metrics include (singleton, sample and statistical metrics):– GE Link Ingress/Egress Average usage– GE Link Ingress Frame Error Ratio– IPv4 Ingress/Egress Packet discard ratio

› These TN metrics can help identify:– under-dimensioned GE links (i.e. highly utilised links)– transport problems/errors over GE links (high frame discard ratios)– IPv4 packet discard issues (queue capacity, errored packets, etc)

Transport Network Performance Metrics (44/1553-HSC 105 50/1)

L2 observability

L3 observability

Slide title 30 pt

Text 18 pt



Transport Network performance

› Additional TN Performance Management counters are available

› SCTP (Signalling transport for S1AP and X2AP)– Sent and received data/control chunks– Dropped chunks (buffer overflows)– Checksum errors

› Synchronization (SoIP related counters)– Provides the highest delay variation counters for the active IP sync reference– Calculated in terms of the best x percentage sync frames experienced during a 100

second window (result in microseconds)– The percentages include 1, 10, 50 %

› IpInterface (Signalling, payload and sync)– Failed Pings to default routers (RPS)– Discards, header and IP address errors

Slide title 30 pt

Text 18 pt



Transport Network tracing

› RPS– $ appdh info / $ appdh rps <ip inter>– Use Wireshark to see ICMP

› QCI, ARP, AMBR values– te e bus_send bus_receive S1AP_ASN

(e.g. InitialContextSetupRequest)

› SCTP– $ te e all cpxSctpIC– $ te e all Scc_SctpHost_proc

› Synchronization (SoIP using NTP)– te e trace7 NSS_CBM_TUM2_TUREG– $ nssinfo all

Slide title 30 pt

Text 18 pt



Transport Summary

› LTE RAN transport topology is very flexible– a mixture of L2/L3 topologies could be implemented (could complicate analysis)

› Basic transport network redundancy is provided (in terms of L3 routers)

› Transport network dimensioning should be taken into account as statistical gains are used (backhaul peaks should be known)

› QoS (Radio and Transport) is essential for proper network operation and should be implemented throughout the network (i.e. in L2/L3 nodes as well)

› Basic performance management is provided in initial LTE RAN releases

› Additional observability can be secured through protocol analysis


44 pt


E2E Analysis

Slide title 30 pt

Text 18 pt



e2e analysis

› End-2-end analysis of throughput is required when end users report throughput issues that are not readily seen in the LTE RAN

› End user throughput investigations/analysis is done at the UE/Server side

› Typical user payload will use the TCP transport protocol

› Layer 3 IP configuration aspects are not covered

› The UDP transport protocol is not covered

Slide title 30 pt

Text 18 pt



lte end user Protocol stack

Telnet, FTP, TFTP, HTTP, SNMP, …..

BGP RIPPort Number

Application Layer

OSPF EGP TCP UDP IMCP IGMPTransport Layer

ARP IP RARPInternet Layer

Protocol Number

Type Code

PDCP, GTP-UData Link Layer

Slide title 30 pt

Text 18 pt



TCP at a glance

› TCP will be the most used transport layer protocol (email, ftp, browsing, etc)› TCP offers the following features:

– Stream data transfer› TCP offers a contiguous stream of segments for applications

– Reliability› Sequence numbers used by sender that expects positive acks (or retransmit)› Receiver uses sequence numbers to rearrange segments (remove duplication)

– Flow control› Receiver indicates the number of bytes it can receive (to sender)

– Multiplexing› Achieved through the use of ports

– Logical connections› Each connection is identified by the pair of sockets used (in receiver & sender)

– Full duplex operation› TCP provides concurrent data streams in both directions

Slide title 30 pt

Text 18 pt



TCP Congestion ControlC

on

ge

stio

n W

ind

ow

3rd DUPACK

3rd DUPACK

3rd DUPACK

Time out

Pipe Capacity

Slow start threshold reached

t

ssthresh on congestion

Fast recovery

Fast retransmit

Congestion Avoidance

Slow Start

Slide title 30 pt

Text 18 pt



TCP Performance

› Bandwidth Delay Product (BDP) - data required to fill the TCP pipe:– Bandwidth of link (bytes per sec) * Delay (sec) = amount of data in transit to fill pipe

› Example: A T1 (1.5 Mbps) over a satellite connection with RTT = 500ms– (1,500,000 bits per sec / 8 bits per byte) * (0.5 sec) = 93,750 bytes– With a rwnd of 65,535, performance is 65,535/93,750 = ~70% or 1.05 Mbps– Hence, the BDP requires more transit data, and this is achieved with an increase in

the rwnd (the TCP scaling feature, RFC1323, can provide this increase)

› TCP Windowing (rwnd) and RTT limit the achievable throughput as follows:–

› Hence, large receive windows and small RTT are desiredTypical LTE RTT

Slide title 30 pt

Text 18 pt



TCP Tuning (Client side)

› TCP performs differently on different Operating Systems (there have been many variations on the TCP congestion control algorithm for instance)

› On Windows Vista:– autotuning = highlyrestricted

› netsh interface tcp set global autotuninglevel=highlyrestricted– autotuninglevel = restricted

› netsh interface tcp set global autotuninglevel=restricted

› On Windows XP:– SP2+: follow this Windows XP TCP configuration KO– Pre-SP2: Use the DrTCP application

› On Linux (depends on the kernel)– Follow this TCP Tuning Guide for kernels 2.4 -> 2.6 (i.e. 2.6.20+ are covered)

Slide title 30 pt

Text 18 pt



LTE RAN TCP Behaviour

› Processing (handovers, buffering, delays, scheduling, retransmissions) in the RBS can affect TCP operation

› Affect of incorrect TCP receive window sizes:– Lower than BDP ->

› Packet loss can lead to (retransmissions, dropped in RBS, etc):– TCP retransmissions and delays– Send rate (throughput) reduced up to 50%, then a linear increase until next drop or

max TCP rate reached

› TCP timeouts can lead to:– TCP fallback into slow start

Slide title 30 pt

Text 18 pt



Analysing e2e traffic

› The following section uses the Wireshark application to inspect e2e traffic (i.e. FTP download from UE)

› Two scenarios are used to highlight how Wireshark can perform detailed analysis on network traffic

› The goal is to show how e2e analysis can be performed from a UE perspective

› The following shows analysis of downlink FTP data session towards a UE

Slide title 30 pt

Text 18 pt



Throughput Results

› For DL throughput, select a packet from Server to UE, and select Statistics -> TCP Stream Graph -> Throughput Graph

Scenario A Scenario B

Slide title 30 pt

Text 18 pt



Throughput Results Cont

› Two other ways to quickly determine the UE throughput are via Statistics -> Summary and Statistics -> IO Graphs

Average added for clarity

Scenario A

Slide title 30 pt

Text 18 pt



Time Sequence Graphs

› A great insight into the TCP performance/behaviour is found with the Statistics -> TCP Stream Graph -> Time-Sequence Graph (Stevens)

Scenario A

Scenario B

Slide title 30 pt

Text 18 pt



TCP Flow Graphs

› TCP flows (Statistics -> Flow Graph) are useful in isolating a particular TCP conversation

Scenario B

Slide title 30 pt

Text 18 pt



Additional TCP Analysis

› Wireshark’s Expert Info (Analyse -> Expert Info) provides a better display of uncommon or notable network behaviour:

Scenario B

Slide title 30 pt

Text 18 pt



Analysis Suite

› Wireshark is a powerful tool to inspect e2e network behaviour

› Other tools can be used in conjunction with Wireshark to offer a complete e2e analysis suite:– iperf

› works in client/server modes› allows UDP and TCP payload to be injected into the network› useful to see link capacities with UDP (max DL/UL throughput)› identifies possible TCP bottlenecks

– Netpersec› Offers real-time display of throughput

– IXIA Chariot Endpoint› commercial product to test IP networks› provides automated application level analysis (HTTP, VoIP, etc)› provides detailed results/statistics of performance

Slide title 30 pt

Text 18 pt



e2e Summary

› The main transport layer (OSI L4) protocol used for network services (internet) will be the Transmission Control Protocol (TCP)

› TCP offers end user application reliable data delivery, flow control and good bandwidth utilisation

› The LTE RAN will implement features/functions that provide better inter-work with the TCP protocol

› Wireshark is a powerful open source network protocol analyser that can offer advanced e2e throughput investigations using inbuilt functions

› Additional tools (like iperf) support the complete e2e analysis

Documents

LTE-Optimization