Position Assignment in Digital Cellular Mobile Radio Networks (e.g. GSM)
`derived from Measurements at the Protocol Interface
`
`Ingo Gaspard
`Deutsche Telekom AG, Technologiezentrum
`P.O. Box 100003,64276 Darmstadt, Germany
`Fax: +49-6151-83 4638
`E-mail: gaspard @I fz.telekom.de
`
`Thomas Engel
`Deutsche Telekom MobilNet GmbH
`P.O. Box 300463,53184 Bonn, Germany
`Fax: +49-228-936 1469
`E-mail: engel @I bn.detemobil.de
`
`Abstract - Measurements are carried out at the protocol interface
`between base transceiver station and base station controller unit
`for optimization of the base station subsystem in an existing
`digital mobile network. At this interface it is possible - in
`contrast to measurements at the air interface - to gain mass data
`(e.g. received power, BER, signaling information) with low
`expenditure. The collected data allow however, only for a very
`rough estimation of the terminals’ positions.
`In this paper a method is presented to assign measurement data
`from protocol interface to the related location of the mobile with
`the aid of field strength prediction. The application of the
`position assignment method is demonstrated within the GSM
`network.
`To validate this approach measurements were taken from the
`German GSM D1 net and were compared with exact position
`assignments by means of GPS.
`On the base of
`location assignment methods there are new
`experimental possibilities available, e.g. identification of traffic
`hot spots or areas of bad interference conditions.
`
`I. INTRODUCTION
`
`In today’s digital cellular mobile radio networks, features like
`power control and handover are related to periodic measurements
`of level and quality at the mobile (downlinWforward) and at the
`base station (uplinkheverse) receiver. The measurement values
`and corresponding signaling events of all customer’s calls in a
`specific cell under investigation could be observed by the
`network operator at
`the protocol
`interface between base
`transceiver station and base station controller unit. Statistical
`evaluation of such mass data produced by customer calls and
`collected at the protocol interface is an important aid to optimize
`the base station subsystem parameters in an operating network.
`The almost only drawback is that there is no exact information
`available about the position of the mobile. Position determination
`is limited to the statement ‘lies in’ or ‘lies out’ of the coverage
`bounderies - which are also known only roughly - of the cell
`under investigation.
`If the performance along routes or at selected points has to be
`examined, there are carried out field test runs by special cars
`equipped with GPS for exact position assignment of downlink
`measurement values. In comparison
`to protocol
`interface
`measurements this procedure implies some disadvantages like
`
`no availability of uplink data, very time consuming and costly
`test runs, no mass data for statistical evaluation available.
`By collection of data from several
`test runs and some
`postprocessing for visualization an impression of
`the real
`coverage of the network can be achieved. In Fig. 1 a part of the
`the German D1 GSM network based on
`coverage of
`measurement runs in the area of Bonn, Germany, is shown. In
`this figure different gray shades along the routes correspond to
`coverage by different base station transceivers which are
`indicated by small circles.
`In section I1 the applied algorithms for position assignment of
`protocol data are presented. All position assignment methods are
`based on a comparison of values measured by the mobiles with
`the values predicted by a typical network planning tool. A
`simulation example shows the influence of prediction errors on
`the accuracy of position assignment.
`Section I11 shows the application of one of the proposed position
`assignment methods to the German D1 net as an example for a
`GSM net in operation. The results are compared with GPS based
`position assignment. It can be seen that the proposed position
`assignment is of sufficient accuracy to be useful e.g. in traffic hot
`spot detection in macro cells.
`
`Fig. 1: Visualization example of coverage evaluated by test runs
`in the area of Bonn of the German D1 GSM network
`
`0-7803-3659-3/97 $1 0.00 0 1 997 IEEE
`592
`Authorized licensed use limited to: Kelly Liu. Downloaded on September 28,2023 at 02:49:45 UTC from IEEE Xplore. Restrictions apply.
`
`Page 1 of 5
`
`SAMSUNG EX-1014
`
`

`

`~
`
`11. POSITION ASSIGNMENT ALGORITHMS
`
`During a call each mobile station monitors signal powers of
`surrounding neighboring cells to assist e.g. features like handoff
`[1],[2].These measurements are available at the cell site and can
`be monitored and stored for further evaluation by the network
`operator with the aid of commonly used protocol analyzers.
`from neighboring cells a
`Besides
`the powers
`received
`measurement of the delay for the signals propagating between
`serving base station and mobile station is performed. In GSM
`this quantity is called 'Timing Advance' (TA). It ranges from 0
`to 63 times half of the bit duration of 3 . 9 w , i.e. one step
`these measurements an
`[l]. By
`corresponds
`to 553m
`experimental vector %(ti) is formed for each time step t i . Its
`components are the powers received from the serving cell and
`from the neighboring cells supplemented by one component
`which represents the delay measurement. The received power can
`be reduced by power control of the base station transmitter, so
`the measured received power of the serving cell has to be
`corrected by the actual power control reduction step which is
`also available at the interface &,is. It is assumed in the following
`that % ( t i ) is characteristic for the mobile's position with a
`resolution defined by the time step t i . Because of averaging in
`the receiver fast fading of the measured power is eliminated. In
`Fig. 2 an example of a real measurement taken at the protocol
`interface Abis of the German D1 net is shown. The maximum
`number of monitored neighboring cells in GSM is equal to 6
`what will limit the maximum available information for further
`processing.
`
`by sliding averaging over at least 5 successive measurement
`points. In Fig. 3 the result of the sliding averaging of the
`measurement data from Fig. 2 is shown as an example.
`
`w
`
`0
`
`I
`
`-
`
`
`
`- & - " I
`-
`I
`$
`Urm*1.Pl*WmHs
`
`E
`
`Fig. 3: Measurement example from Abls after sliding average
`In parallel to the evaluation of the measurement vectors %(r,) a
`set of prediction vectors v ( x , y ) is established once for a given
`cell. This set is generated by prediction of received median
`power as a function of position (x,Y) in the area of interest,
`which is at least as large as the coverage area of the observed
`serving cell. At each point of the grid of longitude and latitude
`the predicted powers for serving and neighboring base stations
`are considered as the components of the prediction vectors
`v ( x , y) . In general, in GSM the number of neighboring cells to
`be considered is limited to 32. Thus the maximum number of
`components for each prediction vector is 32 plus one component
`for the serving cell's power. In Fig. 4 a visualization example for
`the predicted received power of the serving cell in a suburban
`area of about 15kmxlOkm evaluated in the following is shown
`(dark gray = high power, light gray = low power).
`
`E
`
`urm IlholU0m.s
`
`Fig. 2: Measurement example taken from &is
`Measurement values are repeated in the GSM every 480msec.
`Fluctuations of the measured quantities resulting from this fine
`time resolution make a further averaging expedient. This
`averaging is dependent on the maximum assumed speed of the
`mobile and of the spatial resolution of the used prediction tool.
`The resolution of the topographical data base used for prediction
`was 5" which is equal in Germany to about 150m in north-south
`and lOOm in east-west direction, respectively. If we assume
`maximum average speeds of 150km/h a mobile stays within an
`element of the meshed grid at least for 2.4sec. Hence,
`it is
`reasonable to do the further averaging of the measured G(ti) data
`
`Fig. 4: Prediction of received power for serving cell 'Bergheim'
`After preparing the measurement vectors and prediction vectors
`the position assignment is performed by searching the predicted
`vector which is the most similar one to the actual measurement
`vectorii(ti). This search has to be performed for each
`measurement vector. The a priori known position of the most
`
`593
`Authorized licensed use limited to: Kelly Liu. Downloaded on September 28,2023 at 02:49:45 UTC from IEEE Xplore. Restrictions apply.
`
`Page 2 of 5
`
`

`

`~
`
`similar prediction vector is assigned to the measurement vector in
`time step ti . Three different similarity measures were evaluated:
`0 correlation coefficient between vectors &(ti) and v’(x, y )
`
`where n denotes the vector component, i denotes time ti and the
`overbar denotes averaging over all components of the vector.
`
`cosine of the angle between vectors &(ti) and C(x,y)
`
`The position of the predicted vector which yields maximum
`correlation coefficient or maximum cosine of the angle between
`measurement vector and prediction vector is assigned to the
`actual measurement vector.
`0 distance of the vectors &(ti> and v’(x, y )
`
`The position of the predicted vector with minimum distance to
`the measured vector is assigned to the actual measurement
`vector.
`The timing advance information is used to limit the number of
`prediction vectors which are compared to the actual measurement
`vector with the aid of the three defined similarity measures.
`By repeating the steps described so far for every measurement
`vector of a complete call one will get the position assignments
`during the entire call and consequently an estimate of the
`mobile’s run.
`
`111. APPLICATION IN GSM
`
`To evaluate the different methods of position assignment
`measurement runs were carried out. As a reference the position
`assignment of a GPS receiver was used, which resulted in an
`accuracy in the order of about 100m. The measurement runs were
`carried out in a suburban area in the vicinity of Cologne,
`Germany. The cell radius was about 1Okm. So the rectangular
`prediction area was chosen to cover the whole cell area and
`consisted of 210x150 prediction vectors laying on a grid of
`5”x5” with 11 components: 10 possible neighbor cells within
`the cell under investigation plus one component for the serving
`cell ‘Bergheim’. In parallel, the data available at the interface
`Abis were stored. It was found out that most exact position
`assignment was obtained by calculation of the second similarity
`measure between measurement and prediction vectors. The
`reason is that this similarity measure is not dependent on the
`absolute measurement or prediction accuracy but only on the
`relation of the different components. For instance, if the gain of
`
`the real receiver antenna is not equal to the assumed value in the
`prediction, it will not affect the angle between measurement and
`prediction vectors, only a linear scaling of the measurement
`vector is performed.
`The measurement runs were performed at moderate mobile
`speeds and it was found reasonable to do some further averaging
`beyond the averaging of the received powers. This second
`averaging concerns simple averaging after position assignment
`and is done in space by independent averaging in north-south and
`east-west direction.
`As an example, Fig. 5 shows the visualization of position
`assignments with the second similarity measure calculation for
`three different calls during our test drives. For each call we were
`driving the same route. The number of measurement vectors for
`each call was in the order of 600. The route was a closed loop
`and is also visualized by GPS position assignment. It can be seen
`from Fig. 5 that the position assignments were reproduced very
`well for the three independent calls. The accuracy in radial
`direction in relation to circles of constant distances from the
`serving cell is much better than in tangential direction. This is
`because the timing advance information contributes to accuracy
`only in radial direction but not in tangential direction.
`To investigate the influence of prediction errors on the accuracy
`in position assignment different prediction errors in the sense of
`varying standard deviation values were simulated. Starting with
`the prediction vector at the position of the same serving cell
`‘Bergheim’ as we carried out our above mentioned measurements
`we generated 1000 vectors each, where the components are
`generated by independent Gaussian processes with means equal
`to the components of the start vector and standard deviation of
`4dB and 8dB, respectively, which are typical values for state-of-
`the-art prediction tools. These sets of 1000 vectors each were
`processed by
`the position assignment method with cosine
`correlation and the frequency of the assignment to a specific
`position on the 5”x5” grid was calculated. Again the results are
`visualized with an underlying map. Fig. 6 shows the result for
`8dB standard deviation, Fig. 7 depicts the result for 4dB standard
`deviation. As assumed with decreasing standard deviation of the
`prediction, the accuracy in position assignment will increase.
`By comparing simulation and measurement and remembering a
`typical standard deviation of the predicted powers in the order of
`8dB the most important influence on position assignment
`accuracy is the accuracy of the predicted powers.
`the
`For further refinement,
`it is reasonable to interpolate
`predicted powers, what can be done by simple 2-dimensional
`interpolation in space. Another improvement can be gained by
`weighting the single position assignments by the related value of
`similarity calculation. Also some weighting of the measured and
`the predicted powers can improve accuracy. This is reasonable
`due to the fact that in general the largest prediction errors can be
`found for large path loss because in this case many reflections or
`path irregularities have to be considered by the prediction tool
`which lead to larger accumulated resulting errors.
`
`594
`Authorized licensed use limited to: Kelly Liu. Downloaded on September 28,2023 at 02:49:45 UTC from IEEE Xplore. Restrictions apply.
`
`Page 3 of 5
`
`

`

`IV. CONCLUSIONS
`
`In this paper we presented different methods for position
`assignment of mass data from measurements at the protocol
`interface in digital cellular mobile radio networks based on a
`comparison of measurement reports on the mobile terminal
`position and prediction of received powers by a planning tool.
`Measurement results taken from the German D1 GSM net were
`shown and compared with GPS position assignment as reference.
`The mo:st important application of position assignment methods
`introduced in this paper is in ‘hot spot’ detection of areas of high
`traffic. This task is of interest especially in existing mobile
`networks with a fast growing number of subscribers, where small
`cells have to be integrated into the existing infrastructure in an
`
`efficient way without the need for prior installing test base
`stations to find out their optimal locations.
`
`REFERENCES
`
`[ 11
`[2]
`
`[3]
`
`[4]
`
`ETSI, “GSM Recommendations, Sene 05“, 1990
`QUALCOMM, “An Overview of the Application of Code Division
`Multiple Access (CDMA) to Digital Cellular Systems and Personal
`Cellular Networks“, May 1992
`Bronstein, I.N., Semendjajew, K.A., “Taschenbuch der Mathematik‘,
`21th. edition, Teubner, Leipzig, 1983 (in German)
`patent pending ,,Verfahren zur Ortszuordnung von MeBdaten
`ausgewahlter FunkkenngroBen eines zellularen Funknetzes“, DeTeMobil
`Deutsche Telekom MobilNet GmbH, Bonn, Oct. 1995 (in German)
`
`Fig. 5: Three different position measurement examples along the same route plotted in light gray, dark gray and black. For reference
`the GPSi position assignment is plotted in gray.
`
`Authorized licensed use limited to: Kelly Liu. Downloaded on September 28,2023 at 02:49:45 UTC from IEEE Xplore. Restrictions apply.
`
`595
`
`Page 4 of 5
`
`

`

`Frequency of
`assignments:
`
`1 to 3
`
`4 to 7 I 8 to 15
`
`1 6 to 31
`
`32 to 6 3
`
`g.t. 6 3
`
`Fig. 6: Simulation result for frequency of position assignment with 8dB standard deviation: correct position marked by
`small circle
`
`Frequency of
`assignments:
`
`1 to 3
`
`4 to 7
`
`8 to 15
`
`16 to 3 1
`
`32 to 6 3
`
`.I g.t. 6 3
`
`Fig. 7: Simulation result for frequency of position assignment with 4dB standard deviation: correct position marked by
`small circle
`
`Authorized licensed use limited to: Kelly Liu. Downloaded on September 28,2023 at 02:49:45 UTC from IEEE Xplore. Restrictions apply.
`
`596
`
`Page 5 of 5
`
`