View the Project on GitHub SapanaChaudhary/Power-Spectrum-Cartography
With Dr. Radha Krishna Ganti, Dr. Balaraman Ravindran, Dr. Bhaskar Ramamurthi and V. Aravind (SDE, RF Coordination,BSNL, Coimbatore).
In this project, starting from the raw GSM data, I have used an empirical model of the environment to first localize users roughly. Following this, I generate power spectrum maps over the given geographical region. The project is implemented in matlab from scratch.
The data for the purpose of this project is RSL(Radio Signaling Link) data recorded by the Abis Interface. Abis interface is the interface between between BTS and BSC as shown in the diagram below:
The data given by BSNL is from the 3rd layer of Abis Interface Protocol stack. They have provided RSL data. RSL stands for Radio Signalling Link (more).
The hardware used for recording this data is provided by Huawei Technologies Co., Ltd. We sincerely thank BSNL, Coimbatore, India for providing us with the data.
The GSM protocol lists 43 different types of messages that are exchanged between BSC and BTS at the Abis Interface. These messages are broadly categorized into messages for RLM, CCM & TRXM and DCM.
Our data contains the following message types, in Uplink and Downlink directions:
Abis, BCCH, BS, BTS, Channel, Connection, Data, Deactivate, Direction, Error, Establish, Handover, Immediate, MS, Measurement, RF, Release, SACCH, SMS
Following is an instance of trace file from the data: (Fields followed by the values taken by those fields)
Index Report Time Interface Type Message Type Direction Location Condition Content
50137 2016-09-30 11:30:11 910 Abis Interface RSL Abis Cell Pack Page Command Downlink Site ID: 12,Cell ID: 5, TRX No.: 0 7F F0 00 01 00 00 01 53 7F F0 00 01 00 00 01 38 00 00 00 18 00 96 FF FF 00 29 00 00 00 00 00 0C AA 18 01 00 0C 05 F4 1A 64 76 85 01
We have data recorded over 16/09/30 2:59:49 PM to 16/09/30 3:20:14 PM.
The obtained raw data, in the form mentioned above, contains a lot more messages and fields(within those messages) than are useful for the purpose of this project. In this section we extract the useful data. The traces corresponding to message type = ‘Measurement Report’ are extracted and dumped into new set of text files. Each of these text files are then given to the code that outputs csv files consisting of the following fields(column-wise) in the same order:
day, month, year, hour, minutes, seconds, millisecond, site_id, cell_id, trx_values, ul_sender_cpu_id, ul_sender_pid ul_receiver_cpu_id, ul_receiver_pid, ul_length, msg_type, handle, crdlc_ccb, mtls_ccb, auc_reserved, data_length, channel_type, time_slot, rxlev_ncell2, rxlev_full_up, rxqual_full_up, rxqual_sub_up, tlv_type_rx_main_level, tlv_type_rx_sub_level, bs_reserved, bs_power_level, ms_power_level, ms_reserved, actual_timing_adv, timing_reserved enhanced_meas_msg_type, ba_used, dtx_used, rx_lev_full_serving_cell, ba_used_3g, meas_valid, rx_lev_sub_serving_cell spare, rx_qual_full_serving_cell, rx_qual_sub_serving_cell, no_n_cell rxlev_ncell_1, bcch_freq_ncell_1, bsic_ncell_1 rxlev_ncell_2, bcch_freq_ncell_2, bsic_ncell_2, rxlev_ncell_3, bcch_freq_ncell_3, bsic_ncell_3, rxlev_ncell_4 bcch_freq_ncell_4, bsic_ncell_4, rxlev_ncell_5, bcch_freq_ncell_5, bsic_ncell_5, rxlev_ncell_6, bcch_freq_ncell_6, bsic_ncell_6
Sectorized data: The Samathur BS(Base Station) has 3 sectors encompassing 65 degrees each. Figure below depicts the antenna locations for Samathur. Sectors are numbered as 139(Sec-1), 140(Sec-2 and 141(Sec-3).
The mat files sec_139, sec_140 and sec_141 contain the complete sector data with timing information.
The cellular data is now available as a 220327*64 matrix contained in sec_139.mat. 64 is the number of fields(occurring in the same order) as listed above. 220327 is the number of traces, each recorded approximately at an interval of 46.7622 ms. This would mean unique user data comes in approximately at every 11th reading.
The data obtained from the step above contains neighbour cell bcch information as numbers called ncell number. These numbers need to be mapped to actual bcch values, independently for each sector. Now, for each sector of the BS, we construct a data matrix that contains the information on (bcch,bsic,cell ID, RSSI, lat, long) for each of the 6 neighbouring BSs, timing advance for that trace and power received by the UE(User Equipment) from Samathur BS (all the fields in this same order). We follow the following steps to get to this data, individually for each sector :
Further statistics of the final data, like different numbers of neighbouring cells for a particular reading, can now be obtained from the data.
Each row of the data matrix obtained above contains information that can be used to localize that UE at a particular position in and around the cell corresponding to Samathur BS. The localization algorithm followed is simple:
Remove all rssi with rxlev ’0’ and ’1’, Remove all co-sectors of samathur Perform localization without removing the shadowing: bs_power in dB = 2.86789556 (60/31 W) Store the conversion table (say RS) from RXLEV to RSSIs Read BS IDs, location data and RSSIs For each row: create list of neighbours with Lat,Long, cumulative RSSI For every neighbour whose inverted RSSI intersects with inverted BS RSSI, find the centroid and localize user there. Localizing user above would involve both radius and direction, quantities which are appropriately determined. If BS RSSI is the only RSSI, invert it and locate the user anywhere on the arc(ideally 120 degrees, here 65 degrees) formed by respective distance.
The RSSIs are inverted based on the Okumura-Hata model. Parameters used for the same are: Frequency = 900 MHz, Average mobile station height = 2.5 m, Average antenna height = 35m, Different path loss correction factors are used depending on experiments with small sized city/open area. The inversion process can be easily understood from the following matlab code:
% inverted okumura hata model
function [d] = dist_from_pathlossindb(lo)
% lu = path loss in dB from small sized city
% lo = path loss in dB from open area
f=900;
hm=2.5; % average mobile station height
hb=35;
% okumara hata model
% d is in meters
% ch =3.2*(log10(11.75*hm))^2-4.97 ; for large cities
ch = 0.8 + (1.1*log10(f)-0.7)*hm - 1.56*log10(f);
lu = lo + 4.78*(log10(f))^2 - 18.33*log10(f) + 40.94 ;
d = 1000*10^((lu - 69.55 - 26.16*log10(f) + 13.82*log10(hb)+ ch)/(44.9-6.55*log10(hb)));
The figure below is how the localization algorithm is working for a particular trace. The red dot is the BS. The blue dot is where the UE is localized. Hollow circles are the locations of neighbouring BSs. The circles around each of the BSs are the inverted distances corresponding to the RSSI received at those BSs. The x and y axes are latitude and longitude, respectively.
Once the users are localized , the power values corresponding to those users are interpolated using scatteredInterpolant matlab function. The function is used to do interpolation on 2D scattered data. We use the default linear interpolation. The generated heat maps over Samathur and neighbouring cells look like the following figure:
A single call from the same user spans over multiple traces. Each of these traces come at an interval of approximately 480 ms. A better way to do power map reconstruction would be to average the power from the same user; rather than localizing him/her at different locations with corresponding powers.
A user is uniquely identified by: site ID + cell ID + Trx No. + Channel number
An example of a call from the same user is the following: Sector 139: trace # 1,9,17 correspond to the same user.
| Site ID | Cell ID | Trx_No | Channel No. | Time Slot | Timing Adv |
|---|---|---|---|---|---|
| 38 | 139 | 3 | 1 | 0 | 1 |
| 38 | 139 | 3 | 1 | 0 | 1 |
| 38 | 139 | 3 | 1 | 0 | 0 |
Localization for each of the traces, with Samathur as origin, results in following (x,y) coordinates in meters:
| x coordinate | y coordinate |
|---|---|
| 286.737848281044 | 469.341460307053 |
| 287.807527516458 | 468.686277913983 |
| 0 | 0 |