radio propagation channel model tuning - mentum

7/25/2019 Radio Propagation Channel Model Tuning - Mentum

1/22

Radio propagation channelRadio propagation channelModel tuning overviewModel tuning overview

29 Sep 2009


2/22

Overview of propagation model used in planetOverview of propagation model used in planet

Input Scanner or CW drive test

Map, projection

Site configurations used (e.g, link budget, GPS)

Planet general model (PGM) Slope based Okumura-Hata type model

CRC- predict4 model Deterministic (i.e., map dependent, instead of survey), physical-optics based model

Universal model (UM) Additional license required

Unmasked and masked version

Unmasked means that antenna correction is done by planet prediction engine, instead ofUM calculation

2


3/22

PGM overviewPGM overview (1)(1)


4/22

PGM overviewPGM overview (2)(2)

Account for FSL

K1 (freq-dependent intercept),

K2 (slope)

K5 (multiplier for effective antenna height)

Effective antenna height gain

BTS ht gain side as K3

MS ht gain as K6

Clutter effect Weight factor K in last 1km to rx

Diffraction

multiply by K4 for non-LOS

Calculation based on Epstein-Peterson method for 3 diffracting edge

Use clutter height evaluate diffraction for non-LOS

Use Clutter separation as distance between last effective diffracting clutter obstruction to rxantenna


5/22

PGM tuningPGM tuning

PGM only compute vertical diffraction

In DU/U environment where horizontal diffraction can be significant,PGM often over-estimate vertical diffraction loss

Compensate with clutter gain

PGM effective where BTS ht >= surrounding clutter

Using AMT

manual

Use Hata for K3 and K5, clutter offset = optimize

Optimize K1, K2 and K4

smart

Optimize K1 to K5 and CAL in one pass

Optional 2nd step

Fix tuned K and using clutter tuner to re-tune CAL or do manual

change


6/22

CRCCRC--predict4predict4

6

Huygen principle(vector summation of secondary

radiation sources)

Map pixel

Clutter effect specified asclear distance and obstacle

height to receiver


7/22

Comparison of PGM and CRC predict4Comparison of PGM and CRC predict4

7

PGM CRC predict

Application Urban to suburban Accurate clutter/terrain maps,secondary radiation sources

Advantage Fast, good for long distance

propagation

Detailed prediction along many

radials

Typical prediction

resolution (No of radial)

720 360

Weakness Needs more CW data to estimate

slope

Easy to tune, since accuracy

dependent on clutter/terrain,NOTDT data

Model Similar to COST231/Hata-

Okumura, slope-based model withvarious K parameters

Deterministic model based on

Physical optics to calculatediffraction over terrain/clutter

Receiver height Different value assigned to eachclutter

All mobile have same height for allclutter class

Auto tune tool Optimize K, clutter absorption/separation

Optimize clutter absorption property


8/22

Before model tuningBefore model tuning

Add new sites

Setup link budget to get correct EIRP (e..g., PA power, pilot %, cable loss, rxantenna gain)

Add combined gain/loss = receiver antenna gain receiver cable loss,

to DL link budget for all sectors

Similar to manually adjust K1 in PGM

If scanner DT is used

Planet uses RSSI as CPICH RSCP/pilot power for CDMA based network

Allocate scanner record to sector and export as survey

If needed, combine multiple scanner log from same sector to 1 log

Create header for each survey data per sector

Filter survey data

Average survey data

Assign filtered/averaged survey data to associated sector


9/22

Model tuning work flowModel tuning work flow (1) import survey(1) import survey

9


10/22

Model tuning work flowModel tuning work flow (2) create header(2) create header

10

Select sectorwhich survey

belongs to

Survey changes color

after header isgenerated

Site configuration is assignedto that survey


11/22

Model tuning work flowModel tuning work flow (3) filter survey(3) filter survey

11

Extract valid survey data formodel tuning


12/22

Model tuning work flowModel tuning work flow (4) average survey(4) average survey

Remove fast Rayleigh fading

10~20

or about 2m for 2.5GHz Average by distance to avoid bias effect or 1/3 of map pixel or 5m

(use 2m since it is smaller)

12


13/22

Model tuning work flowModel tuning work flow (5) assign to sector(5) assign to sector

13

Used to compare prediction coverage with survey

data by calculating mean/std/RMS error


14/22

Survey histogram (after filter and averaging)Survey histogram (after filter and averaging)

Dense urban sites

urban sites

suburban sites

Smooth monotonically rolloff onboth ends of dBm


15/22

Clutter distribution (after filter and averaging)Clutter distribution (after filter and averaging)

urban sites

suburban sites

(recommend 2000~3000 sample per

clutter class for good model tuning,absolute minimum is 200~300 sample

per clutter class)

Dense urban sites


16/22

Distance regression (after filter and averaging)Distance regression (after filter and averaging)

Dense urban sites

urban sites

suburban sites

DU model should has steepest slope(i.e., larger K2 magnitude) compared

to U and SU model


17/22

Model tuning (1)Model tuning (1) createcreate untuneduntuned versionversion

17

Clutter separation ~ 1 or 2 pixel distance, depending on environmentMost clutter have some diffraction loss (except water)

For PGM Diffraction loss is pre-calculated based on clutterseparation/height, only clutter absorption loss is tuned


18/22

Model tuning (2)Model tuning (2) automatic model tuner (PGM)automatic model tuner (PGM)

18

Use smart to tune all K values and CAL in one passOptional 2nd step after running AMT, run CAL tuner to tune

CAL only with fixed K values obtained from AMT


19/22

Model tuning (3)Model tuning (3) verify tuned model (PGM)verify tuned model (PGM)

19

Check model tuning report

K and CAL comparison BEFORE and AFTER

running AMT (automatic model tuner)

Check errorIf negative model error model is over-predicting (i.e.,predicted dBm is higher than survey)

CAL is only calculated if survey available in that clutterclass (if no survey, set to 0 by default)

If clutter separation is too short, diffraction loss calculatedwill be too high.If clutter has gain, it basically means clutter separation is

too low or clutter height too high uses clutter class with most sample as a reference to

compute K1 and compare with other clutter type give -/+clutter absorption loss


20/22


21/22

Model tuningModel tuning (5)(5) sanity check for PGM modelssanity check for PGM models

21

DU site with

PGM DU model

DU sites with

PGM U model

DU site with

PGM SU model


22/22

Model tuningModel tuning (6)(6) predicted coverage with surveypredicted coverage with survey

22

Legends for predicted coverage and

thematic map of survey are same

radio propagation channel model tuning - mentum

Documents