twintrx coverage

All

right

s res

erve

d. P

assi

ng o

n an

d co

pyin

g of

this

do

cum

ent,

use

and

com

mun

icat

ion

of it

s con

tent

s no

t per

mitt

ed w

ithou

t writ

ten

auth

oriz

atio

n fr

om A

lcat

el.

3DF 01902 2910 VAZZA Edition 01 RELEASED 6/32

1 INTRODUCTION

The objective of this document is to provide all necessary information to Radio Network Engineers (RNEs) supporting the introduction of the TwinTRX in coverage mode in a network, and having to demonstrate its gains to a customer. This document will therefore cover the following aspects:

Description of the new TwinTRX module

Principles of the coverage mode: TxDiv and 4RxDiv/TMA features

Recommended configurations on field (antennas, number of Twin modules, etc.)

Principles of coverage gain assessment

Drive tests methodology: pre-requisites, tools, resources, planning

Measurements post-processing methodology and implementation (tool) This document describes the state of the art at the date of publishing, but forthcoming field trials will certainly bring new information on the topic, and should be taken as an opportunity to enrich this method.


2 TWINTRX PRESENTATION

2.1 PRINCIPLE The Twin TRX is the latest generation of TRX issued from a family which evolved as follows:

Figure 1: TRX generations history

Figure 2: TWIN TRX

Twin TRX can operate in two modes: Capacity mode: In this configuration, the Twin TRX is equivalent to 2 TRX MP in one housing. The consequence is that, by maximum, 24 TRX can be set in a single cabinet (this maximum is currently supported by MBI5 functional variant & MBO2 BTSs). Coverage mode: In this configuration, both transmission units are used for a single TRX. Thanks to the "Transmit diversity" (TxDiv) feature, the transmission power is increased (this power increase is environment dependent) and it can reach 175W (in theory) in dense urban environments using GSM900 frequency band.

Figure 3: TWIN TRX modes of operation


The above scheme illustrate the possible ways Twin TRX can be used in a rural environment (rural environment is taken as an example).

All

right

s res

erve

d. P

assi

ng o

n an

d co

pyin

g of

this

do

cum

ent,

use

and

com

mun

icat

ion

of it

s con

tent

s no

t per

mitt

ed w

ithou

t writ

ten

auth

oriz

atio

n fr

om A

lcat

el.

As seen from the coverage mode specifications, equivalent transmitting power in rural environments with 900 MHz frequency band is not 175 W, but rather 113 W ;which proves that TxDiv gains are different from one environment to another (according to scattering and fading conditions). TxDiv gains are explained in details in section 1.1.3. It can also be noted in this example that along with 4Rxdiv solution for UL sensitivity enhancement, TMA can also be a good choice as it is preferred by some operators which have restrictions in antenna installation. TWIN TRX module benefits from:

• Flexibility as it may works in 2 modes: capacity or coverage • Decreased footprint and weight in capacity mode • Lower number of sites in coverage mode • Power consumption reduced up to 35% per TRX

2.2 CAPACITY MODE

In this configuration, one Twin module is equivalent to 2 functional TRX: 16 RTS. Both TRX have the same radio performances as EDGE+ TRX MP:

TRX of a Twin TRX unit, set in capacity mode, can belong to different sectors.

Thanks to the new BTS SW, the cabinets can support up to 24 TRX (note that the maximum number of TRXs per cell is still limited to 16 TRXs).

Already installed Evolium BTS can welcome twin modules: mixtures of generations of TRX and antenna networks are supported.

The capacity mode will not be detailed in this document, as it presents no difference with a G4 medium power TRX, from a radio engineering point of view.

2.3 COVERAGE MODE

The purpose of this mode is to increase the coverage area. Both transmission units are used for one single functional TRX. As a consequence, only 8 TS are available per twin TRX module, and only 12 TRX can be supported by a single cabinet.

Several associated features are linked to this mode in order to:

Increase the transmission power (by using TxDiv)

Increase the sensitivity at BTS level (UL direction). This is mandatory to keep a well-balanced link budget. (by using 2-Rx+TMA or 4-Rx div according to radio configs)

Thanks to these radio signal combinations, the radio performances are as follows:


2.3.1 PRINCIPLE OF TXDIV The TxDiv consists in two elements: - Use of both transmission units of the twin TRX module. The same signal is transmitted twice on two different antennas. This leads to an increase of 3dB of the output power (On-air combining).

Figure 4: On-air combining with Txdiv

But the coherence between those signals can lead to destructive effects. That’s why a delay between both signals has been introduced (See below). A 2-symbol delay is introduced between the transmission paths.

Figure 5: Transmission delay impact on fading reduction

This delay allows reducing the risk of simultaneous strong attenuation of signal on the both channels. A high de-correlation between the signals is introduced, thus the diversity gain is increased. As we can see on the graph, the fading holes are often compensated by the other path’s signal. A trade-off had to be found because if the introduced delay is too long, it can lead to self-interferences phenomenon. Gain according to this de correlation is up to 2.9 dB depending on the environment.


Summary of Tx div gains:

All

right

s res

erve

d. P

assi

ng o

n an

d co

pyin

g of

this

do

cum

ent,

use

and

com

mun

icat

ion

of it

s con

tent

s no

t per

mitt

ed w

ithou

t writ

ten

auth

oriz

atio

n fr

om A

lcat

el.

The diversity gain is maximal in urban environments because of the high signal scattering and more multi-paths. On the contrary, in rural environments, the gain is less.

2.3.2 PRINCIPLE OF 4RXDIV The principle is to receive the signal from 4 different antennas, and to combine it into the Twin TRX module. That’s why 4Rx div is only available with Twin TRX in coverage mode. On top of 2-Rx div, this brings an additional gain. This gain is different according to the radio environment:

• Dense urban: 4 dB • Suburban: 3.6 dB • Rural: 2.9 dB

The prerequisite is that 4 different reception paths exist on the sector (4 vertical polarization antennas, or 2 cross-polarization antennas). The aim of this feature is to improve UL sensitivity, and reject a strong interferer, as for the 2Rx div. Two main algorithms are used:

• Optimum ratio: method used to combine signals and take benefit from the diversity gain.

• Least square: beam-forming algorithm to reject a strong interferer. Summary of 4Rx div gains:


2.3.3 TMA USAGE (TOWER MOUNTED AMPLIFIER)

TMA can be used as an alternate option to 4Rx div feature. However, it should be noted that TMA introduces additional losses in DL path; this is called TMA insertion loss (typically 0.4-0.6dB).

This solution can be adopted in case operators don't want to add additional antenna to apply 4Rxdiv. It also suites the cases where the use of additional ANCs (or AGCs) is restricted – sometimes because the maximum number of feeder entrances supported by current BTS cabinet is already reached with 2Rx configuration – so using TMA makes it possible to use the same BTS cabinet with no need of cabinet extension. However, TMA can be useless in some cases where using one ANC with TWIN means that additional combining losses will be introduced, which may cancel the Txdiv gain. Thus, using TMA with 2 ANCs per sector (and accordingly 4 vpol/2xpol antennas) has no meaning as 4Rxdiv should be used instead. Using Friess formula, TMA contribution can be calculated so that: Appendix A it can be compared to theoretical 4Rxdiv sensitivity gain corresponding to the radio environment of concern Appendix B Link budget calculations should be made to ensure path balance with TMA. Please refer to [2] for more information about Friess formula.

3 RECOMMENDED CONFIGURATIONS ON FIELD

3.1 REMINDER ON ANTENNA NETWORKS

The figure below presents the antenna network used nowadays in most BTS, the ANC. It replaces the previously used combination of ANX (duplexer) and ANY (combiner / splitter).

The latest evolution of the ANC is the AGC, which is slightly more compact (1/4 of subrack instead of 1/3). AGC is not mandatory with TwinTRX, except for higher configurations in capacity mode (to save some space in the cabinet). Therefore small configurations with Twin TRX in coverage mode can use standard ANC.

As a reminder, the ANC can be used in low-loss mode (with by-pass function activated) when only 1 or 2 TRX at the maximum are connected. In such case, the ANC loss is only 0.8 dB, to be compared to the loss of 4.2 dB when ANC is not in low-loss mode.

Antenna A

TXA - RXA - RXdivB

SplitterWBC

TRX 1TXRXnRXd

TRX 2TXRXnRXd

Splitter

Splitter

LNA

Duplexer

FilterFilter

Splitter Splitter WBC

Antenna BTXB- RXB - RXdivA

Duplexer

FilterFilter

Splitter

LNA

n

Antenna ATXA - RXA - RXdivB

SplitterSplitterWBCWBC

TRX 1TXRXnRXd

TRX 1TXRXnRXd

TRX 2TXRXnRXd

TRX 2TXRXnRXd

SplitterSplitter

SplitterSplitter

LNALNA

Duplexer

FilterFilter

Duplexer

FilterFilter

SplitterSplitter SplitterSplitter WBCWBC


Duplexer

FilterFilter

Duplexer

FilterFilter

SplitterSplitter

LNA

n


SplitterWBC

TRX 1TXRXnRXd

TRX 2TXRXnRXd

TRX 3TXRXnRXd

TRX 4TXRXnRXd

Splitter

Splitter

LNA

Duplexer

FilterFilter

Splitter Splitter WBC


Duplexer

FilterFilter

Splitter

LNA


SplitterSplitterWBCWBC

TRX 1TXRXnRXd

TRX 1TXRXnRXd

TRX 2TXRXnRXd

TRX 2TXRXnRXd

TRX 3TXRXnRXd

TRX 3TXRXnRXd

TRX 4TXRXnRXd

TRX 4TXRXnRXd

SplitterSplitter

SplitterSplitter

LNALNA

Duplexer

FilterFilter

Duplexer

FilterFilter

SplitterSplitter SplitterSplitter WBCWBC


Duplexer

FilterFilter

Duplexer

FilterFilter

SplitterSplitter

LNA

Figure 7: Antenna Network Combiner (ANC) in normal mode (non by-passed)

Figure 6: Antenna Network Combiner (ANC) in low-loss mode (by-passed)


3.2 DESCRIPTION OF THE RECOMMENDED CONFIGURATIONS

When defining the configurations to be tested in coverage mode, with the customer, two important points shall be kept in mind:

• The Twin TRX in coverage mode requires double antenna paths compared to a normal TRX. • The ANC shall be kept in by-pass mode, because the additional combining loss would cancel the

coverage gain brought by the TxDiv

All r

ight

s re

serv

ed.

Pass

ing

on a

nd c

opyi

ng o

f th

is

docu

men

t, u

se a

nd c

omm

unic

atio

n of

its

cont

ents

no

t pe

rmit

ted

wit

hout

wri

tten

aut

hori

zati

on f

rom

Alc

atel

.

As a consequence, some configurations shall be avoided, including:

• 2 Twin TRX in coverage mode with only 1 ANC • or 4 Twin TRX with one in coverage mode (except if Unbalanced TRX output power is used, see

below)

The configurations described here below are the advised ones, offering maximum coverage to the operator.

3.2.1 CONFIGURATIONS WITH 1 TRX PER SECTOR

The “1 TRX” configuration is particular because it offers 2 possibilities in UL, to balance the link budget and the increase of output power due to TxDiv:

• Either the 4RxDiv feature

• Either a TMA

The TMA sensitivity gain is similar to the 4RxDiv (around 3 to 4dB), provided that the feeder length is high enough. Its gain can be computed thanks to the TMA technical characteristics and feeder loss, with the Friess formula. Thanks also to the new TMA generations, only 1 TMA is needed for 2 vertically polarized antennas or 1 cross polarized antenna.

If TMA is used with AGC monitoring, TMA parameters must be retrieved from manufacturer, and entered at BTS side during commissioning:

• TMA threshold low

• TMA threshold high

• Antenna feeder loss

• TMA Tx loss

• TMA Rx gain

Figure 8: Typical configuration with 1 TRX



Note on TMA: with 2 Twin TRX in coverage mode, the number of Tx paths needed (4) requires 2 ANCs, and therefore 4 antennas (otherwise, the ANCs would need to be in combined mode, with an additional loss of 3.4 dB which is not acceptable).

In that case the 4RxDiv features comes “for free”, and brings an UL sensitivity gain similar to the TMA. That is why TMA is not an efficient solution for such configurations and above.

ANC

G4

TRX

TWIN

TRX

A1 A2

G4

TRX

ANC

A1 A2 A3 A4

ANC

TWIN

TRX

G4 TRX inlow-loss mode

TwinTRXwith TxDiv+4RxDiv

ANCANC

G4

TRX

TWIN

TRX

A1 A2A1 A2

G4

TRX

ANCANC

A1 A2A1 A2 A3 A4A3 A4

ANCANC

TWIN

TRX


TwinTRXwith TxDiv+4RxDiv



For configurations with 3 TRXs and above, with coverage mode, need to make use of another feature, called “Unbalanced TRX Output Power”. This feature comes with B9MR1, and has been tested successfully in Orange France during B9MR4 pilot.

Its principle is to allow a difference of output power between different TRX in the same cell (removing the automatic leveling). To do this, the cell is configured as concentric, with an output power higher on the outer zone than the inner zone.

The figure below presents the configurations to be applied in that case:

The coverage mode is ensured by a Twin TRX cabled on the ANCs in low-loss mode. It is mapped on the outer zone.

The 2 other TRX, handling the extra traffic, can be either 2 G4 TRX or one Twin TRX in capacity mode. They have additional loss, and are mapped on the inner zone.


ANC

G4

TRX

TWIN

TRX

A1 A2

G4

TRX

G4

TRX

A3 A4

TWIN

TRX

ANC ANC

A1 A2 A3 A4

ANC


1 TwinTRX in coverage mode (TxDiv+4RxDiv)1 TwinTRX in capacity modeUnbalanced TRX Output Power activated

ANCANC

G4

TRX

TWIN

TRX

A1 A2A1 A2

G4

TRX

G4

TRX

A3 A4A3 A4

TWIN

TRX

ANCANC ANCANC

A1 A2A1 A2 A3 A4A3 A4

ANCANC



All

right

s res

erve

d. P

assi

ng o

n an

d co

pyin

g of

this

do

cum

ent,

use

and

com

mun

icat

ion

of it

s con

tent

s no

t per

mitt

ed w

ithou

t writ

ten

auth

oriz

atio

n fr

om A

lcat

el.


3.2.4 CONFIGURATIONS WITH 4 TRX OR MORE

The configuration with 4 TRX is very similar to the one above, with 3 TRX. It needs to make use of “Unbalanced TRX Output Power”, in order to combine both constraints of coverage mode with highest possible output power, and the need of several antenna paths for capacity.

TWIN

TRX

ANC

G4

TRX

TWIN

TRX

A1 A2

G4

TRX

G4

TRX

A3 A4

G4

TRX

TWIN

TRX

ANC ANC

A1 A2 A3 A4

ANC



TWIN

TRX

ANCANC

G4

TRX

TWIN

TRX

A1 A2A1 A2

G4

TRX

G4

TRX

A3 A4A3 A4

G4

TRX

TWIN

TRX

ANCANC ANCANC

A1 A2A1 A2 A3 A4A3 A4

ANCANC





4 ASSESSMENT OF DL AND UL SENSITIVITY GAINS

4.1 PRINCIPLE OF DL EQUIVALENT OUTPUT POWER GAIN ASSESSMENT (TXDIV)

To assess Txdiv gain, drive tests should be carried out for each configuration (with & without Txdiv feature activation).

Blue: RxLevDL for G4 HPRed: RxLevDL for TwinTRX TxDiv

In DL, the RxLev is measured directly by the Trace mobile (for instance, Sagem OT290 with Agilent Nitro drive test tool). It is assumed that this RxLev takes into account the global TxDiv gain: on-air combining of the signal at the output of BTS antenna and additional diversity gain linked to slight time shifting at Tx.

RxLevDL is simply binned on a square grid, and then averaged (in dBm) on the drive test route.

The averaged RxLevDL got from different tested configurations (with & without Txdiv) are compared together to get Txdiv gain. Figure 12: RxLevDL plot

Figure 12 shows RxLevDL at each geographical location using G4 HP TRX and TWIN TRX with Txdiv.

Note: Mapinfo can be used for general performance judgement but not for precise gain calculation.

Why is binning needed? In order to get accurate results when comparing the averaged RxLevDL from different drive tests, binning should be applied for the following reasons:

- It is impossible to have exactly the same number of samples in all drive test versions. - Measurements taken from one drive test to another are not passing exactly (due to also the

inaccuracy of GPS) by the same points; and thus inaccurate results can be expected. - The car used for drive test may not have exactly the same speed for all DTs (drive tests), so there is

a probability to have in one DT a low speed near the site (thus generating a lot of samples in range of good level) and higher speed in another DT (thus generating less number of samples with good level).

- Being stationary (stopping for example due to traffic lights) at certain places can have a similar effect as the car speed; for example if we stop at a region of bad radio conditions during the tests, a lot of samples with bad RxLevDL will be recorded and thus degrading the whole RxLevDL average of the route, which doesn't reflect the real behaviour.

- Drive test measurements are subjected to changes due to having (for instance) a truck passing near the DT car which will cause changes in the measurements in the range of 3 dB.

So a solution should be found in order to compare different drive tests despite their inexactness. Square Binning is simply averaging all samples that lay in a square grid (typical grid size = 20m) and thus resulting in one sample per square grid. This sample location can be either in the square grid centre or in the average location of all DT samples lying inside the square grid). The binned samples over square grids are then averaged over the whole DT route to get the final average RxLevDL. Thanks to binning, this average RxLevDL can be then compared for different drive tests. With binning, results reliability is guaranteed. Figure 13 illustrates the square binning idea.


All

right

s res

erve

d. P

assi

ng o

n an

d co

pyin

g of

this

do

cum

ent,

use

and

com

mun

icat

ion

of it

s con

tent

s no

t per

mitt

ed w

ithou

t writ

ten

auth

oriz

atio

n fr

om A

lcat

el.

Figure 13: Binning on a square grid

Important note:

Drive test route should be scanned several times to ensure sufficient number of samples and increase the accuracy of the binning.

Before After

Figure 14: Samples before & after binning

The graph below is a plot of RxLev DL for TwinTRX in coverage mode and G4 MP TRX, done in TWIN TRX trial in Kenya.

Figure 15: DL Received Level with G4 TRX and Twin TRX


Average gain for each RxLev range is computed by taking the average RxLevDL (after binning and taking only the common points between G4 and Twin traces) over the whole DT route (after filtering in excel on the required RxLevDL range). So there will be 2 RxLevDL averages: one for G4 and the other for Twin. Thus, the average gain is the difference between the RxLevDL_Twin and RxLevDL_G4.

4.2 PRINCIPLE OF UL SENSITIVITY GAIN ASSESSMENT (4RXDIV OR TMA)

4.2.1 INTRODUCTION Assessment of the gain of the TxDiv is quite easy to perform, as measurements of RxLev and RxQual in DL are provided by the drive test tool chain, along with GPS positioning. Also, it’s assumed that the RxLevDL reported by the trace mobile reflects the gain from the TxDiv, after combining of the different signals received on the single antenna. Regarding 4RxDiv gain (compared to 2RxDiv), the gain assessment is more difficult to perform, for several reasons:

1. The RxLevUL measurements are indeed taken from the Abis, and therefore not linked to GPS coordinates.

2. Moreover, RxLevUL on the Abis does not reflect the full 4 RxDiv gain, as it reports only the value of the best signal of 4 antennas (in case of 4Rxdiv) or the signal amplification due to the power amplifier of the TMA which may be in the order of 12-15 dB power gain (if TMA is used). More details about UL RxLev reporting are shown in section 3.2.3.

Hence, the goal of the coming sections is to detail the possible methods to be used to assess the gain of this 4RxDiv feature.

4.2.2 INPUTS

Drive tests have been done with different configurations: G4 2RxDiv, Twin 2RxDiv+TMA, Twin 4RxDiv MS trace provides Timestamp, GPS position, RxLevDL, RxQualDL Abis trace provides Timestamp, RxLevUL, RxQualUL

How to get GPS position for UL measurements?

In order to get the GPS position for UL measurements, the UL and DL traces can be correlated using timestamps along with some filtering steps on abis trace (to get the measurements corresponding to DT test call).

Usually, there is a small time shift between clocks of the PC recording DL measurements through Agilent

Nitro and K12 (or K15) protocol analyzer recording abis measurements. So the time-shift between both clocks needs to be estimated in order to get accurate synchronization of time between the two traces (DT trace & Abis trace). This timeshift estimation can be done using "CONNECT" message recorded in both traces.

4.2.3 BTS MEASUREMENTS REPORTED ON ABIS FOR RXLEVUL According to BTS-CC, the RxLevUL value reported by the BTS is the one for the best antenna, not for the combination of 4 antennas. With 4Rxdiv feature, there are 4 antennas (either 2 cross polarized antennas or 4 vertically polarized antennas). Each antenna receives a different RxLevUL and the 4 RxLevUL values got from the 4 antennas are then combined together to get an equivalent received level that is better than all of the 4 signals received by the four antennas. However, the BTS doesn't report this equivalent RxLev (after combining), but it only reports the best received signal among the 4 signals coming to the 4 antennas.


So it is clear that the RxLevUL value reported by the BTS is slightly higher with 4Rxdiv than with 2Rxdiv due to the higher probability to get a better signal level with 4 receiving paths than with 2 receiving paths. This gain resulting from having 4 antennas instead of 2 is typically in the range of 1dB but it is also clear that it doesn't completely reflect the total sensitivity gain of the 4Rxdiv feature.

All

right

s res

erve

d. P

assi

ng o

n an

d co

pyin

g of

this

do

cum

ent,

use

and

com

mun

icat

ion

of it

s con

tent

s no

t per

mitt

ed w

ithou

t writ

ten

auth

oriz

atio

n fr

om A

lcat

el.

The true gain can be estimated from signal to noise ratio (SNR) improvement and thus considering RxQualUL (not RxLevUL) as it fully reflects the complete sensitivity gain. As a numerical example; let’s assume that the received signals before the combining of the demodulator are: Ant1: -80 dBm, Ant2: -82 dBm, Ant3: -79 dBm, Ant4: -83 dBm Then the demodulator reports to the decoder the max value of the 4 antennas for each burst => this means antenna 3 for this example: the decoder averages for one multiframe (in linear scale) and sends this value to the Abis interface. The combiner inside the demodulator generates one signal out of these 4 signals (combining burst by burst, not only selecting bursts) => Combined signal strength around -75 dBm. But this value is not reported. The reported value corresponds only to the air power.

4.3 PRINCIPLE OF UL SENSITIVITY GAIN ASSESSMENT (4RXDIV OR TMA)

4.3.1 METHOD1: ERICSSON METHOD This is the first method that was proposed: a) For each drive route, compute the average of RxQualUL, for a given RxLevUL. b) From the graph obtained, assess the RxLevUL threshold for which we have a degradation of RxQual (for instance RxQual > 1).

Figure 16: Sensitivity assessment: RxQual vs. RxLev

Numerical example on a sample DT: -> G4_2Rx: -106.3 dBm -> Twin_2Rx: -108.9 dBm -> Twin_4Rx: -108.7 dBm c) Compute the average RxLevUL on the drive route, to assess the average improvement of the 4Rx (compared to 2Rx), due to measurement of "best Rx on 4-ways" instead of "best Rx on 2-ways". Assuming following RxLevUL values for the sample DT: -> G4_2Rx: -95.5 dBm -> Twin_2Rx: -95.9 dBm


-> Twin_4Rx: -94.6 dBm d) When comparing Twin_4Rx and G4_2Rx, or Twin_4Rx and Twin_2Rx, correct the threshold computed in b), with the "artificial" RxLev gain computed in c). As a conclusion from the sample DT: -> Global gain from Twin_4Rx compared to G4_2Rx: 2.4dB + 0.9 dB = 3.3 dB -> Global gain from Twin_4Rx compared to Twin_2Rx: -0.2dB + 1.3 dB = 1.1 dB -> Global gain from Twin_2Rx compared to G4_2Rx: 2.6dB – 0.4 dB = 2.2 dB Conclusions: -> In this method, the delta in RxLev between 2 Rx and 4Rx is averaged on the whole drive test route, and then applied to the RxLev threshold for which we have bad RxQual. This could lead to very approximate results. Remark: Nevertheless, this method has been used by Ericsson and presented in its whitepaper “Field Trial Results of 4-Way Receive Diversity in a Live GSM network”, available at the following URL: http://www.ericsson.com/technology/research_papers/atsp/papers/field_trial_results.shtml

4.3.2 METHOD2: ALCATEL-LUCENT METHOD In the second method, it’s proposed to assess directly the RxLev threshold for which we have a degradation of the RxQual, without the differences of RxLev measurement between 2Rx and 4Rx. So we need to use the same RxLevUL as a reference for a given point (for example the RxLevUL measured with the G4_2Rx), and retrieve RxQualUL for each configuration, in order to get the distribution of samples RxQualUL / RxLevUL for each case, with the same RxLevUL reference. In order to do this, we need to:

a) Get the RxLevUL and RxQualUL for the reference (G4_2Rx for example) in a given geographical point (GPS position) b) Get the RxQualUL for the Twin_4Rx at the same position c) Produce the two distribution graphs of RxQualUL_4Rx / RxLevUL_ref, and RxQualUL_ref / RxLevUL_ref d) Measure between the 2 graphs the difference of thresholds of RxLevUL for a given degradation of RxQualUL (for example RxQual > 1). However to do this, we need to retrieve GPS positions for UL samples, and then correlate the samples at a given GPS position between two successive drive routes, knowing that GPS positions are never exactly the same from one drive route to another. To solve this issue, square binning is proposed as done in Txdiv gain assessment: measurements on a GPS position on a grid are averaged together. This also solves the issue related to traffic lights (when from one drive route to another, more samples are retrieved at a bad level, simply because the car stopped longer at a given bad position due to a traffic light).

4.3.3 METHOD3: COVERAGE EXTENSION METHOD As the methods above can be quite difficult to put into practice, and are not very "customer-friendly", we propose a simple method, more oriented towards customer demonstration purpose: -> Ensure with link budget calculation that we are in a configuration UL limited -> Drive a route till the coverage limit of the cell, with the reference configuration -> Drive a route till the coverage limit of the cell, with the 4RxDiv configuration -> Check that the coverage with conf2 is extended compared to conf1

http://www.ericsson.com/technology/research_papers/atsp/papers/field_trial_results.shtml


4.3.4 METHOD4: POWER REDUCTION METHOD As a variant of Method3: -> Ensure with link budget calculation that we are in a configuration UL limited -> Drive a route till the coverage limit of the cell, with the reference configuration -> Decrease the MS transmission power by the value of the theoretical gain expected from 4RxDiv

All

right

s res

erve

d. P

assi

ng o

n an

d co

pyin

g of

this

do

cum

ent,

use

and

com

mun

icat

ion

of it

s con

tent

s no

t per

mitt

ed w

ithou

t writ

ten

auth

oriz

atio

n fr

om A

lcat

el.

-> Drive again the same route till the coverage limit of the cell, with the 4RxDiv configuration -> Check that the coverage with conf2 is the same as for conf1

4.3.5 EXAMPLE The goal of this example is to illustrate the application of the Alcatel-Lucent method.

Point GPS1 Point GPS2RxLevUL_Ant1 = -94 RxLevUL_Ant1 = -97RxLevUL_Ant2 = -96 RxLevUL_Ant2 = -99

RxLevUL_Abis = -94 RxLevUL_Abis = -97RxQualUL_Abis = 2 RxQualUL_Abis = 4

DT with 2 RxDiv

Point GPS1 Point GPS2RxLevUL_Ant1 = -94 RxLevUL_Ant1 = -97RxLevUL_Ant2 = -96 RxLevUL_Ant2 = -99RxLevUL_Ant3 = -93 RxLevUL_Ant3 = -96RxLevUL_Ant4 = -97 RxLevUL_Ant4 = -100RxLevUL_Abis = -93 RxLevUL_Abis = -96RxQualUL_Abis = 1 RxQualUL_Abis = 2

DT with 4 RxDiv

As seen from this example, the "best antenna" computation used for RxLevUL_Abis shows 1dB gain, but it is not the global gain from the 4RxDiv algorithm, just the fact that with 4 antennas there are more chances to have good signal than with 2 antennas. On the other hand, with 4RxDiv powerful algorithms, we expect to have a big improvement in the resulting RxQual at a given geographical point, by combining 4 signals instead of 2. So the Alcatel-Lucent method is the following:

a) Trace the RxLev_2Rx – RxQual_2Rx, to see that RxQual=2 for RxLev_UL = -94dBm. b) Trace RxLev_2Rx – RxQual_4Rx, to see that the new point for RxQual=2 is now a little further, actually 3

dB lower if you consider the same RxLev measurement as a reference. So we should be able to conclude from this analysis that the global gain of 4RxDiv is 3 dB.


Figure 17: RxLev/RxQual graph

Reminder on Ericsson method:

a) First we do RxLev1-RxQual1, to conclude that the threshold for RxQual=2 is RxLev_Abis=-94dBm in 2RxDiv.

b) Then we do RxLev2-RxQual2, to conclude that the threshold for RxQual=2 is RxLev_Abis=-96dBm in 4RxDiv.

c) Then we compare the averages of RxLev between 2 RxDiv and 4 RxDiv to conclude that the artificial gain of "best antenna" is around 1dB in average

So finally we conclude that the global gain of 4RxDiv is 3 dB. In this example, the result is the same with Method1 than Method2, but Method2 is more accurate in case "best antenna" gain is not equal in each point of the drive route.

4.4 CONCLUSION Method2 is the preferred one, as considered the most accurate, to evaluate the actual gain of the 4RxDiv feature on a given drive test route. An implementation of this method has been developed in a quick & light tool by Network Engineering (see next section). In case of a need to demonstrate such gains to a customer, Method3 or 4 can be used, as they are easier to set up.

All

right

s res

erve

d. P

assi

ng o

n an

d co

pyin

g of

this

do

cum

ent,

use

and

com

mun

icat

ion

of it

s con

tent

s no

t per

mitt

ed w

ithou

t writ

ten

auth

oriz

atio

n fr

om A

lcat

el.


5 DRIVE TESTS METHODOLOGY

5.1 TOOLS

The following tools must be used:

1) Drive test toolchain, including: Trace MS PC Trace software (preferably Agilent Nitro E6474A, as our post-processing tool is adapted to it), including

proper license/dongle GPS Power supply for PC and MS Connecting cables (PC to MS, etc.) Rooftop antenna (car kit)

2) Tektronix K12 or K15 with proper cables and Abis decoding stacks 3) One PC for offline post-processing, with following software installed:

Microsoft Access and Excel (2003 version is recommended) Drive test software (Agilent Nitro or TEMS). Usually no license is required for offline processing of traces. K12/K15 record viewer with proper Abis decoding stacks MapInfo (including raster map of the area) Network Engineering Quick&Light post-processing software (latest version)

4) NPA/RNO toolchain, for QoS follow-up. RMS jobs shall be activated. 5) Last but not the least, a very important pre-requisite for drive tests: one phone number shall be

available (in some cases with a special SIM card), to call the MSC and therefore have automatic pickup and unlimited call duration (looping voice message)

5.2 RESOURCES

Setting up a TwinTRX field trial requires the following resources: 2 RNEs (1 for drive tests follow-up and post-processing, and 1 for RNO follow-up) 1 drive tester 1 van + driver 1 BSS technician for OMC and K12 operations 1 BSS technician for on-site installation and commissioning Installation teams for antennas, TMAs, etc. on the tower

5.3 PARAMETER SETTINGS

DOWNLINK_DTX_ENABLE = 0 (DTX not allowed) DTX_INDICATOR = 2 (MS shall not use Uplink DTX) EN_BS_PC = 0 (Disable downlink power control) EN_MS_PC = 0 (Disable uplink power control) Half-rate disabled Disabling of DTX, power control and HO blocking (by disabling HO and reselection inside MS trace mobile) allows having a more accurate vision of the coverage gain by staying locked on the tested cell, and signal strength fluctuations are avoided. HR shall be disabled, in order to be able to filter the right call on Abis TS, for 4Rx diversity assessment.


Important note:

A clean frequency should be assigned to test TWIN in coverage mode, in order to avoid radio coverage limitation caused by interference.

5.4 PRACTICAL RECOMMENDATIONS In order to increase binning accuracy, it is recommended to keep the same drive test route and conditions (day time, traffic, weather…) for all drive tests performed during the test. The route should span most of the cell coverage area (reaching cell border). Abis trace should be performed at the same time of the drive test. It is very important to accurately reproduce each time the same drive route, including driving at same speed, on the same lane of the road. Mask effects from other vehicles shall be avoided as much as possible (do not stay just behind or just in front of a big truck for instance). Of course, it is also recommended to maintain a slow and steady driving speed. For each configuration, it is highly recommended to repeat several times the drive test route, if possible on the same day, without any modification. The goal is to increase the number of measurement samples for each geographical location.

6 IMPLEMENTATION OF POST-PROCESSING METHOD IN A TOOL: "NE POST-PROCESSING TOOL"

6.1 INTRODUCTION During the first field trials, which took place on Stuttgart and Velizy on-air platforms, then in Kenya and Algeria, the accurate assessment of the gain appeared to be a much more complex task than expected. Indeed from one drive test to another, on the same configuration, the average of measured RxLev on the drive tests route was fluctuating by several dB, making it impossible to assess the gain with accuracy. The main identified reason for this is that one drive test route can never be reproduced twice exactly the same way (same number of measurement samples exactly at the same geographical position). Therefore the need to perform a precise binning has been identified, and implemented in a tool. As discussed in details in section 3.1, this binning means that all measurements done at a given geographical position (in fact, a 20m square on a GPS grid) are averaged together. This implementation was done in our own tool, because similar functions in other tools (Agilent, Ranopt, etc.) were not giving satisfying results, and we could not know exactly their implementation. A few other functions have also been implemented in our tool:

- the synchronization of UL and DL traces, in order to get the UL measurements at a given GPS position (for further binning, and comparison between configurations)

- the filtering of own measurements in K12 UL traces (as several commercial MS can be present) - the filtering of worst samples, as on some drive tests, the measurements can be suddenly far worse than the

other times, due to sudden fading or masking effects


Figure 18: Effect of fading/masking and need for filtering

All

right

s res

erve

d. P

assi

ng o

n an

d co

pyin

g of

this

do

cum

ent,

use

and

com

mun

icat

ion

of it

s con

tent

s no

t per

mitt

ed w

ithou

t writ

ten

auth

oriz

atio

n fr

om A

lcat

el.

6.2 PREPARATION OF THE FILES In a first step, the trace files obtained in Agilent Nitro and K12/K15 Tektronics need to be prepared and exported (assuming 3 DTs were carried out for each configuration):

Figure 19: input files preparation diagram

Nitro file .SD5 N°1 K12 file .RF5 N°1



Agilent Nitro + Export Plan

(TwinTRX_Nitro.epf)

Data_Nitro.csv

K12 file 1+2+3.rf5

K12 Viewer + Proper Abis stacks + filter TwinTRX_K12.flt

+ columns def TwinTRX_K12.mcc

K12 Merger

Data_K12.csv


The export plan "TwinTRX_Nitro.epf", K12 filter "TwinTRX_K12.flt" and K12 column definition file "TWINTRX_K12.mcc" are all available with the NE post-processing tool package. Output of different Nitro files (.sd5 files) corresponding to different DT versions done for the same H/W configuration can be merged by selecting "Multiple sd5 files exporting (Batch processing)" in the export wizard from tools menu inside Agilent Nitro software, then use single exported file. The same can be done with K12 trace files (.rf5 files) that can be merged together into a single .rf5 file using RecordFileMerger.exe existing in tools directory from K12 application. This merge allows dealing with a single exported file for Nitro traces (Data_Nitro.csv) and another single exported file for K12 traces (Data_K12.csv) to be introduced to NE post-processing tool instead of dealing with many files corresponding to many drive test versions. This solution may be under restriction if the time shift between Nitro PC and K12 machine is not the same for all DT versions (see next section for details). In case TEMS is used for drive tests instead of Agilent Nitro, we will have a file called Data_TEMS instead of Data_Nitro. A file " TwinTRX_TEMS.tex” is also available with the tool package to deal with TEMS exports. To note that in this case, time shift should be manually computed in order to be introduced to the tool.

6.3 ASSESS TIME SHIFT BETWEEN CLOCKS OF NITRO AND K12 In a second step, as the clock of Agilent Nitro PC and the one of the K12 are most probably not exactly synchronized, the time shift between the two must be assessed:

Figure 20: time shift assessment

The export plan "TwinTRX_Nitro_SETUP.epf" is different from "TwinTRX_Nitro.epf" mentioned in figure 19; it is used to filter on setup messages . It is also available with the NE post-processing tool package.

Principle:


a) Open Data_Nitro_SETUP.csv and filter on IEI=773 (this is the SETUP message). Note down the timestamp.

b) Then open Data_K12_SETUP.csv (which has been generated with good stacks, filter and column setup in K12 record viewer) and check the timestamp of SETUP message with the good called number.

Time Primitive IEI Type Msg 05/02/2007 13:18:56.683 3 773 GSM CC SETUP 05/02/2007 13:18:57.595 3 770 GSM CC CPROC 05/02/2007 13:18:58.515 3 771 GSM CC PROGRES

Agilent Nitro + Export Plan

(TwinTRX_Nitro_SETUP.epf)

K12 Viewer + Proper Abis stacks + filter (TwinTRX_K12_SETUP.flt)

+ columns def (TwinTRX_K12_SETUP.mcc)

Data_Nitro_SETUP.csv Data_K12_SETUP.csv


05/02/2007 13:18:58.877 3 771 GSM CC ALERT 05/02/2007 13:18:59.037 3 775 GSM CC CONNECT 05/02/2007 13:18:59.046 3 783 GSM CC CONACK

Example of Data_Nitro_SETUP.csv

All

right

s res

erve

d. P

assi

ng o

n an

d co

pyin

g of

this

do

cum

ent,

use

and

com

mun

icat

ion

of it

s con

tent

s no

t per

mitt

ed w

ithou

t writ

ten

auth

oriz

atio

n fr

om A

lcat

el.

Short Time Last MSG Called party number 13:25:50 SETUP `0734460482` 13:26:10 SETUP `0736258275` 13:26:10 SETUP `0734194879` 13:26:15 SETUP 13:26:22 SETUP `0735928712`

Example of Data_K12_SETUP.csv

In this example, timeshift is: 13:26:10 - 13:18:57 = 433 sec

6.4 POST-PROCESSING IN MSACCESS TOOL In a third step, the files obtained can be put in the same directory as the tool, and this one can be run, through the macros provided, in several steps:

1) Import the files using the corresponding macros 2) Fill the required options in the table “Options” 3) Run the post-processing with the dedicated macro (depending on whether you use Nitro or TEMS for

DL measurements, and whether or not you also want to post-process UL measurements)

Figure 21: Macros provided in the tool

6.5 USING OUTPUT TABLES FROM THE TOOL

6.5.1 OUTPUT RESULTS FROM THE TOOL To date, the MsAccess tool automates only part of the process for gain assessment:

binning (on geographical grid) and filtering (worst samples) of DL data synchronization of UL and DL traces + filtering on test MS in UL trace binning (on geographical grid) and filtering (worst samples) of UL data

As a result, the tool provides 2 tables (result_UL_K12 for UL and result_DL_AIR for DL), each containing the following information:

RxLev (average RxLev of all samples in the given grid square) RxQual (average RxQual of all samples in the given grid square) AvgTA (average TA of all samples in the given grid square)


AvgNorth (avg Northing GPS position, UTM format, of all samples in the given grid square) AvgEast (avg Easting GPS position, UTM format, of all samples in the given grid square) GridNorth (Northing GPS position, UTM format, of the given grid square) GridEast (Easting GPS position, UTM format, of the given grid square) MinOfTime (timestamp of 1st sample on the given grid square, for “sorting” purpose)

Hence the global process for gain assessment is the following: 1) Run the tool

Run the tool with one set of traces, for a reference configuration Rename the tables obtained in MsAccess tool, with “_REF” at the end Run again the tool with a 2nd set of traces, for the test configuration Rename the tables obtained in MsAccess tool, with “_TEST” at the end

2) Process DL measurements (see after), by comparing the 2 tables for DL 3) Process UL measurements (see after), by comparing the 2 tables for UL

6.5.2 PROCESSING DL MEASUREMENTS

For DL gain assessment, build a query comparing RxLev between tables result_DL_AIR_REF and table result_DL_AIR_TEST, based on same Grid points (make 2 links respectively between GridNorth and GridEast of each table)

The result obtained is a table providing the average RxLev measured, for each configuration, at the

same geographical coordinates (i.e. on the same drive test points). Then it can be used either to plot the results in Mapinfo, either to compute a global average on the drive test route and get the average gain.

6.5.3 PROCESSING UL MEASUREMENTS For UL gain assessment, in order to apply the Alcatel-Lucent method (described in § Error! Reference source not found.), you have to follow several steps:

Build a query as described below, comparing RxQual between table result_UL_K12_REF and table result_UL_K12_TEST, based on same Grid points (make 2 links respectively between GridNorth and GridEast of each table). This query will also get the RxLev from the reference DT, round it to the next integer, and group the GPS points having the same RxLev together, averaging the RxQual measured.

Figure 22: Query correlating UL traces of 2 configurations by common GPS coordinates


The result should look like this:

All

right

s res

erve

d. P

assi

ng o

n an

d co

pyin

g of

this

do

cum

ent,

use

and

com

mun

icat

ion

of it

s con

tent

s no

t per

mitt

ed w

ithou

t writ

ten

auth

oriz

atio

n fr

om A

lcat

el.

Figure 23: Query output

Such results can then be copy/paste to Excel.

It should be noted that RxLev value of -111dB corresponds to RxLevUL reported by the BTS having value as "less than -110dB". The BTS reports RxLevUL levels between -47 and -110, so the levels less than -110dB are mapped inside NE post-processing tool to -111dB.

The trend can be computed by Excel, with “Polynomial order 6” function. It is advised to do it on RxLev from –85 dBm to –111 dBm, but removing the RxQual samples measured for RxLev=-111dBm (just keep blank values in Excel), because such RxLev also includes lower RxLev (-112, -113 and below) which the BTS can report only as –111, and therefore the averaged RxQual for such RxLev value is not accurate.

Last but no the least, the difference of RxLev obtained for the 2 configurations, at a given RxQual threshold (chosen arbitrarily as RxQual=1) reflects the UL sensitivity gain.

Figure 24: RxLev/RxQual UL plot


7 ANNEX: DESCRIPTION OF THE MSACCESS TOOL INTERNALS The figure presented here summarizes all the steps done by the tool, and they are detailed just after.

Figure 25: Internal processing done by NE tool

9 6

8 5

4

2

7

3

1

Data_TEMS.csv Data_Nitro.csv Data_K12.csv

DegtoUTM conversion

procedu

Import_Nitro()

Data_Nitro

qNitro

Options

Data_K12

Rename

qAbis2

Data_ul_or_dl Data_ul_or_dl

qBin1

qBin2

qBin3

qBin4

qBin__RUN

OR

Options

result_ul_or_dl result_ul_or_dl

OR

Rename

Result_DL_Nitro Result_UL_K12

tmp_Nitro_or_TEM

Import_K12 ()

Import_TEMS()

Data_TEMS

qTEMS

qAbis_RUN

qAbis1

Rename

All

right

s res

erve

d. P

assi

ng o

n an

d co

pyin

g of

this

do

cum

ent,

use

and

com

mun

icat

ion

of it

s con

tent

s no

t per

mitt

ed w

ithou

t writ

ten

auth

oriz

atio

n fr

om A

lcat

el.


Before entering the description of each step, please note that this tool can be used either to post-process both DL and UL data (in that case, all steps in the diagram below, from 1) to 9) are done successively), or to post-process only DL data (in that case, only steps 1), 3), 7) 8) and 9) are done successively). NB: it is not possible to post-process only UL data, as they need to use DL traces to get the GPS position. Description of all steps, from 1) to 9): 1) Import Data_Nitro.csv:

Clean the table Data_Nitro in Access (to make it empty)

Imports the file Data_Nitro.csv into Data_Nitro table, with DL measurements and GPS position. To do this, we use the specification “Data_Nitro_Import_Spec”, which is accessible for modification in the menu File -> Get External Data -> Import -> Advanced -> Specifications

2) Import Data_K12.csv:

Clean the table Data_K12 in Access (to make it empty)

Imports the file Data_K12.csv into Data_K12 table, with UL measurements. To do this, we use the specification “Data_K12_Import_Spec”

Before next step, there is a part to be done manually:

i. The user shall enter the tables Data_Nitro and Data_K12 in MsAccess (double-click) to check that they are correctly imported. Is they are empty, it means that the source file was not located, or not at the proper format.

ii. Inside the tables imported, the user can check the values to be used for Option table. In DL: ARFCN, and in UL: TEI (usually you should take the one which appears most often, if most of the traffic captured on the BTS is the one from your tests)

iii. You can then modify the 2 values accordingly in Option table before moving to next step.

3) Query “qNitro” run on Data_Nitro table

Filters on correct values in all fields (not empty, not out of range, etc.)

Filters on DL samples with good ARFCN value (taken in Option table)

Takes only the 11 right characters of the timestamp provided by Nitro (the result is the number of the day + time, i-e looks like “19 11:54:09”)

Removes the 5 last characters from the GPS coordinates (9812920.11553(37M))

Puts the result in table tmp_Nitro_or_TEMS

4) Query qAbis_RUN is launched with 2 tables as inputs: tmp_Nitro_or_TEMS and Data_K12

NB: this query calls first two other queries: qAbis_1 and qAbis_2

qAbis_1 performs the following steps, on data_K12 table

i. Convert RxLev from text to integer, thanks to table __conversion_RxLev

ii. Convert RxQual from text to integer, thanks to table __conversion_RxQual

iii. Filters on good TEI, with the value taken in Option table

iv. Generates the good timestamp (same format as above) by applying some conversion of format + shift of time from the number of seconds provided in Option table

v. Groups measurements having the same timestamp (with precision of 1sec). If several of them are found for one given second, averages of RxLev, RxQual and TA are taken.


qAbis_2 performs the following steps, on tmp_Nitro_or_TEMS table

i. In the table tmp_Nitro_or_TEMS, it groups measurements having the same timestamp (with precision of 1sec). If several of them are found for one given second, averages of the GPS coordinates and TA are taken. The goal is just to retrieve the GPS coordinates of the MS at each second.

qAbis_RUN performs the following steps, on the results from qAbis1 and qAbis2

i. For each given second, it matches the samples in DL and UL traces, provided that they have the same TS number, and that their TA is also similar (equal + or – 2)

ii. The result is the same as in qAbis1 (i-e we have the RxLev, RxQual and TA values in UL for each given second) + the GPS coordinates from the MS at the same given second, taken from the DL trace

iii. The result is put in a table called data_ul_or_dl, containing UL measurements with GPS coordinates, ready for binning

5) Query qBin_RUN is launched on table data_ul_or_dl

NB: this query can run the same way on DL or UL data; because they have exactly the same format once they are pre-processed and put in table data_ul_or_dl

qBin_RUN is run after successively: qBin_1, qBin_2, qBin_3 and qBin_4 (each one takes the previous one as input)

qBin_1 performs the following steps, on data_ul_or_dl table:

i. Takes GPS coordinates expressed in UTM (which is a distance in meters from an origin point) then divides it by GridSize (expressed in meters, in table Options), then takes the integer and re-multiply it by the same value.

ii. As a result, we get coordinates GridNorth and GridEasth that are averaged on a grid, which size is provided by Gridsize.

qBin_2 performs the following steps, on qBin_1 results:

i. Computes a new field called “TimeFrame”, which is the Time rounded in minutes (only the 8 most left characters are kept).

ii. For each given set of GPS coordinates on the grid, and each different value of TimeFrame (given minute) it takes the average of RxLev, RxQual and TA samples. It also takes the average of real GPS coordinates, and the first real timestamp.

iii. For each GPS position + timeframe, it counts also the number of samples.

iv. The idea is to differentiate the measurements taken at different period of time (different drive tests, or same DT but to differentiate the way to go, and the way to come back) on a same location, in order to filter out the worst set of measurements in the next queries.

qBin_4 performs the following steps, on qBin_3 results:

i. Filters out all GPS location where we do not have enough samples to do the filtering of worst samples (if WorstSpls=1, then we need at least 2 different measurements on a GPS location to remove the worst one).

ii. NB: if “worst samples filtering” is deactivated (i-e WorstSpls=0), then this step does nothing

qBin_RUN performs the following steps, on qBin_4 results:

i. Effective only if “worst samples filtering” is activated (i-e WorstSpls=1),

ii. For each GPS location, it takes the average of RxLev, multiplies it by the number of times we drove at this place (number of timeframes), removes the worst one, then divide by the number of timeframes minus 1.

iii. It also does the same with RxQual.

twintrx coverage

Documents