Radio propagation channel Model tuning overview
29 Sep 2009
Overview 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 Okumura-Hata type model model
• CR CRCC- predi predict4 ct4 mod model el – 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 of UM calculation 2
PGM overview – (1)
PGM overview – (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 rx antenna
PGM 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
CRC--predict4 CRC
Clutter effect specified as clear distance and obstacle height to receiver Map pixel
Huygen principle (vector summation of secondary radiation sources)
6
Comparison of PGM and CRC predict4 PGM
7
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, NOT DT data
Model
Similar to COST231/HataOkumura, slope-based model with various K parameters
Deterministic model based on Physical optics to calculate diffraction over terrain/clutter
Receiver height
Different value assigned to each clutter
All mobile have same height for all clutter class
Auto tune tool
Optimize K, clutter absorption/ separation
Optimize clutter absorption property
Before model tuning • Add new sites – Setup link budget to get correct EIRP (e..g., PA power, pilot %, cable loss, rx antenna 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
Model tuning work flow – (1) import survey
9
Model tuning work flow – (2) create header Site configuration is assigned to that survey
Select sector which survey belongs to
10
Survey changes color after header is generated
Model tuning work flow – (3) filter survey
Extract valid survey data for model tuning
11
Model tuning work flow – (4) average survey
• Remove fast Rayleigh fading
à
10~20 λ or about 2m for 2.5GHz
• Average by distance to avoid bias effect (use 2m since it is smaller) 12
à
½ or 1/3 of map pixel or 5m
Model tuning work flow – (5) assign to sector
13
Used to compare prediction coverage with survey data by calculating mean/std/RMS error
Survey histogram (after filter and averaging) Dense urban sites
Smooth monotonically rolloff on both ends of dBm
urban sites
suburban sites
Clutter distribution (after filter and averaging)
Dense urban sites
urban sites (recommend 2000~3000 sample per clutter class for good model tuning, absolute minimum is 200~300 sample per clutter class)
suburban sites
Distance regression (after filter and averaging) Dense urban sites DU model should has steepest slope (i.e., larger K2 magnitude) compared to U and SU model
urban sites
suburban sites
Model tuning (1) – create untuned version
•Clutter separation ~ 1 or 2 pixel distance, depending on environment •Most clutter have some diffraction loss (except water) •For PGM à Diffraction loss is pre-calculated based on clutter separation/height, only clutter absorption loss is tuned 17
Model tuning (2) – – automatic model tuner (PGM)
•Use smart to tune all K values and CAL in one pass •Optional 2nd step à after running AMT, run CAL tuner to tune CAL only with fixed K values obtained from AMT 18
Model tuning (3) – – verify tuned model (PGM) Check model tuning report K and CAL comparison BEFORE and AFTER running AMT (automatic model tuner)
Check error •If negative model error à model is over-predicting (i.e., predicted dBm is higher than survey) •CAL is only calculated if survey available in that clutter class (if no survey, set to 0 by default) •If clutter separation is too short, diffraction loss calculated will 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 à give -/+ clutter absorption loss
19
Model tuning (4) – – check error between survey vs prediction based on tuned model Rerun prediction using tuned model, and check error for each sector
Rule of thumb à <3dB mean, <9dB std
20
Model tuning (5) – – sanity check for PGM models DU site with PGM DU model
21
DU sites with PGM U model
DU site with PGM SU model
Model tuning (6) – – predicted coverage with survey Legends for predicted coverage and thematic map of survey are same
22
