NOVEL TECHNIQUE FOR CO-CHANNEL INTERFERENCE MEASUREMENTS IN CELLULAR NETWORKS

Transcription

1 NOVEL TECHNIQUE FOR CO-CHANNEL INTERFERENCE MEASUREMENTS IN CELLULAR NETWORKS Sergey Dickey DTI, a PCTEL company Germantown, Maryland ABSTRACT A novel approach for accurate area propagation measurements in presence of severe CCI is described. Cochannel interference is an inherent characteristic of cellular wireless networks. Existing frequency-planning and optimization tools rely on the validity of path loss data. These data are obtained by drive testing the networks or by using propagation models. Existing drive testing approaches to interference measurements rely on the decoding of base station identification codes (aka color codes) and are not efficient in the areas where the measurements are most critical, i.e., at the edge of cells coverage. The method and its implementation, as well as some results, are described in some detail 1. INTRODUCTION Co-channel interference (CCI) is inherent in any cellular network. This makes any measurement of RF coverage of individual sectors in a TDMA network a painful task. The lack of efficient measurement techniques leads to a reliance by network planners on propagation prediction models, which are renowned for their lack of accuracy, especially in urban environments where the most daunting optimization problems arise. The new measurement technique developed by DTI (patents are pending) and described here allows to accurately measure signal levels of individual sectors in presence of severe CCI. The increased area where signal coverage is measurable increases the efficiency of drive testing significantly, since every measurement point now contains the results for multiple sectors as contrasted with previous measurements which contained just one sector or no identifiable data. The ease of the measurement that leaves any subjective interpretation or guess work out may lead to a new emphasis on drive testing and improved performance of post-processing network planning and optimization tools. 2. BACKGROUND Drive testing is an essential element of the network planning and optimization process. During a drive testing campaign, network planners utilize vehicles equipped with test handsets and in many cases fast-scanning measurement receivers, a GPS receiver for geolocation, a notebook PC for controlling equipment and data logging, in order to separately map signal levels received from multiple sectors in a wireless network. Ideally, the collected maps should describe the actual propagation from sectors and be input to the providers' optimization software tools such as automatic frequency planners, simulators, and network planning tools. In reality, however, drive testing faces a decreasing role due to its inability to effectively resolve signals that originated from different co-channel sectors in the same network (adjacent-channel interference is a lesser problem). Instead, network planners rely more on propagation models and switch statistics, which makes automated network optimization more art than science and in some cases only marginally useful. The most common technique used for coverage testing involves decoding the so-called "color" codes embedded in the signal. In the case of GSM, this code is called BSIC (base station identification code) and is contained in the synchronization burst (SCH). Any measurement point which had a color code decoded is assigned to the nearest sector known to originate this particular color code. The described method works well for the areas with relatively low levels of CCI, namely where signal to interference ratio (SIR) exceeds 2 4 db, but it yields only marginal results in cases where the CIR is close to zero or slightly negative (+/- 2 db), which is characteristic of a cell's periphery, and it does not work at all inside neighbor cells. These are the areas which are vital for the needs of network optimization. In absence of the color code information, the existing drive test tools resign to attempts to attribute measurement points to the originating sectors based on the proximity, antenna pattern, prediction model, etc. In other

2 words, these are guesses that favor neighbors over more distant sectors and do not account for unusual propagation conditions caused by the terrain "cingularities" (hills, lakes, marshes, etc.), elevated roadways and many other reasons. One particular technique of this "signal association" developed during the last several years, relies on the statistical frequencies of various color codes' detection in order to come to estimates of relative contributions of several base stations to the total power received at a measurement point. This works, to a degree, but we hear continuous complaints about the scarcity of the usable measurement points out of the total number of points (no more than %, for the most part) and about lack of such points outside of the serving cell boundaries. A more radical approach consists of temporary re-tuning of the network during the drive test which removes most of the potential sources of CCI from the area to be measured, at the RF frequency of the sector of interest. This is a laborious method, since it requires multiple re-drives, and it creates many disparate views of the network that the post-processing tools have to deal with and manage. 3. AREA MEASUREMENTS VERSUS POINT MEASUREMENTS Given the described in the previous section deplorable state of signal coverage measurements in TDMA networks, solving this technical difficulty apparently became the Holy Grail of drive testing tool development. Since signal association has been performed through color code decoding, it was only natural that one of the attempts at solving the problem focused on simultaneous (joint) decoding of two BSICs. [1]. But however straightforward this approach might be, its realization was extremely difficult. The reason is that BSIC is little different from any other piece of digital information in the signal: it is coded convolutionally bit by bit, and there is no additional processing gain (redundancy) associated with it. Decoding data in presence of interference requires a good knowledge of each of the partial communication channels (for signal and each of the important interferers) and is difficult to obtain in a dynamic environment without requiring huge processing power. The approach above is an example of what can be called "point measurements". At every point, a complete decoding of the digital information and point association are attempted. If we realize that all BSIC codes in an area where a particular signal is present are identical, we can see how redundant this approach is if it is possible to find another characteristic of the signal that allows to distinguish it from other signals in the mixture, and if this other characteristic have a sensible processing gain associated with it. In this case, it is sufficient to uniquely identify the originating sector of the signal component only once (in the worst case) and then only connect other measurement points with the identified points using the common characteristic of the signal instances from the same sector. We will call this latter approach "area measurements". The area-measurement idea is the foundation of the method that is the subject of this discussion. 4. RF SIGNATURES AND FADING One can recognize individual signals in presence of CCI by their "RF signatures", or, in other words, known patterns in the signal with at least one parameter differing between signals from different stations. Among these signatures one can count training sequences and synchronization bursts. Training sequences can be distinguished by their code and time of arrival; the synchronization bursts only by their time of arrival. As example, Fig. 1 shows the result of the correlation of a real-life GSM BCCH signal with the theoretical FCCH ("Frequency-Correction Burst") pattern in the radio receiver. The FCCH burst is a string of 156 zeros in the GMSK signal. Every triangular correlation peak in Fig. 1 corresponds to a distinct signal component in the receiver input bandwidth. As is seen in this figure, they arrive at different times. Figure 1: Input signal correlated with FCCH pattern Fig. 1 is quite instructive: it shows several features of the correlated signal that indicate certain problems and benefits of using this particular burst for decomposing the signal. First of all, the processing gain, and consequently the measurement dynamic range, of the FCCH burst is the best provided by any of single bursts in the signal, since the FCCH is the longest fixed pattern. Secondly, any string of zero bits in the convolutionally-coded data

3 stream will yield a correlation peak of varying duration and amplitude; these can be seen as low-level noise "thicket" at around db under the highest peak (i.e., the peak when the CIR is high). Unfortunately, the GSM data structure contains periodic sequences of zeroes that may result in a regular pattern after correlation; this depends on the relative timing between the receiver's digitizer and the signal, but the pattern, which is convenient to call "ghosting", can be sorted out by the receiver. Use of the FCCH for the signal identification and relative power measurement allows simplifying the receiver s complexity and increasing scanning speed. A very reasonable measurement dynamic range of about 12 to 16 db can be achieved by acquiring only one instance of the FCCH burst. From the GSM standard [2] follows that the FCCH has a true period of 51 frames of the BCCH signal and a "quasi" period of only 10 frames; in other words, it is enough to acquire only 11 frames to guarantee the capture of at least one FCCH burst in every co-channel component of the signal. In some rare instances timing alone will not suffice to resolve separate signal components: this will happen when the originating stations' timings are too close. In these cases use of the SCH mid-ambles, with a much narrower autocorrelation function will help; if the times of arrival are coinciding completely, only the use of training sequences can do the job, since there are eight distinct sequences that are not likely to be the same in interfering signals. Cross-correlation properties of the eight sequences are not too good. Fig. 1 demonstrates that the FCCH correlation peaks have a well-recognized shape. They are easily recognized by a human eye, but for a machine we need to use a "matched" filter to recognize its shape and maximize performance. Another filter is needed to suppress the effects of fading that are slight, but still visible in Fig. 1; Fig. 2 shows a clearer picture of fading. Figure 2: Fading signal Fig. 2 demonstrates that without the use of a "defading" band-pass filter after correlation, one cannot detect some of the correlation peaks, as well as in some cases the varying "baseline" will create patterns that could be detected as false correlation peaks. In both figures, the relative height of the peak corresponds to the relative power of the corresponding cochannel component of the signal 5. OFF-LINE PROCESSING The task of identifying co-channel components by the source sector is performed off-line, after the data from the scanning receiver have been imported into a data base. The fortunate circumstance that facilitates the correlation peak identification is that most if not all BTSs in the market are getting their timing clock from the same core network (typically, deriving it from the connecting T1 lines). Consequently, the correlation peaks from different sectors in the network will have constant mutual delays for long periods of time (propagation delays are small in comparison with frame timing differences). In our experience, the delays were not noticeably changing for days and weeks, although the relative timing between the receiver time base and the network clock is randomly shifted in each of the drives and is slowly drifting during the duration of the drive. The drifts are compensated for by a tracking timing filter that corrects timing of individual measurement points separately during each of the drives. The following major steps comprise the off-line data processing algorithm Correlation peak classification This step further can be subdivided into several sub-steps, but the main idea is that first the peaks of the time-of-arrival distribution are linked (associated) to the sectors by using the most relevant measurement points, typically taken close to the cell sites in question. The information needed for this classification includes the distribution of BSIC points, signal strengths, point location relative to the location and orientation of the sector antennae, etc. The additional sub-steps include the analysis of the resulting table of sector delays, linking together delays for separate drives, and finding closely-spaced delay between different sectors on each of the BCCH channels. If any irresolvable timing relationship was found, an attempt to use a different kind of RF signature is made (i.e., use of a training sequence or another well-defined pattern with a narrow auto-correlation function). The timing table is generated for all BCCH channels that have been scanned.

4 5.2. Timing assignment verification Timing assignment table is checked for consistency between individual drives; 10/11 frame uncertainties are resolved using frame number parameter from the receiver Measurement point attribution Each of the measurement points in the data base is passed through a system of logical filters where it is analyzed for its timing versus drive number. Each of the filters is a timing window with the center time taken from the timing table from the row corresponding to the drive during which the given measurement point was collected. Each such logical filter corresponds to an individual sector in the network, and the points that passed such a filter form the coverage map for this sector. Depending on the relative strength of the correlation peak, or more accurately, its number in the sorted list of returned peaks for each measurement point, a different threshold may be applied in the filter in order to optimize the ratio of the number of detected points to the number of "false alarms" Map verification Each of the sector maps created as described in the previous paragraph undergoes additional checks that serve to cross-check individual points as well as the timing assignments that generated subsets of such points. Among the tools used for cross-checking points there are searches for points that evidently are clustered in "alien" cells (cells surrounding other co-channel sites and having a preponderance of points with a "wrong" BSIC value). In addition, checks of spatial point frequencies and level versus distance profiles can be used at this stage Data export and mapping tool The export tool allows the user to create an output file in DTI standardized format (an Access.mdb file with tables inside containing maps with new and "traditional", i.e., RSSI/BSIC views of the coverage). For the new-style maps, each table contains the map for one of the sectors of the network; there are as many tables as there were sectors in the drive-test area. The number of the old-style maps is equal to the number of BCCH channels scanned. The exported data can be read by a number of post-processing network-planning tools from the companies with whom DTI has signed partnering agreements. These tools allow the user to automatically perform network frequency planning, troubleshooting and network optimization. In addition, the same exporting tool has mapping capabilities and can be used for simple troubleshooting and data verification before using more specialized tools. It allows the user to view RF power coverage and C/I maps for individual sectors as well as RSSI and BSIC plots for any frequency channel. 6. IMPLEMENTATION: CLARIFY TM SYSTEM 6.1. Scanning receiver The scanning receiver uses its DSP engine to perform the following tasks, among others: - data acquisition; - BSIC decoding; - signal power measurements with a number of calibrations; - correlation with the RF signatures; associated timing and power calculations and calibrations; - correlation peak detection; - peak amplitude sorting. The receiver has another microprocessor for control and messaging, as well as a synchronization circuitry and a built-in GPS receiver. GPS lock during scanning is recommended, but not absolutely necessary, so that the system can be used in tunnels and indoors. The receiver communicates with the host (a notebook PC) via a serial link. The host runs datacollection software with control, monitoring, mapping, and data-logging capabilities Data analyzer and viewer/exporter This is a software package built around a SQL Server data base. It can be installed on a remote server as well as on a local machine. All code has been written in VC RESULTS VALIDATION 7.1. Trials and validation techniques During the course of this development, multiple trials with a number of carriers and service providers, both domestic and international, have been carried out. The data were presented to the network RF planning groups for inspection. In addition, several "troubled" spots were identified and presented that were characterized by high levels of CCI. One of the trials included a direct comparison of the results obtained in the presence of severe CCI with those for the same locations, but measured before two of the sectors in a small area had been re-tuned in order to create this high level of CCI. Overall, the drive-test area of a few square miles, characterized by moderately-sized hills and stream valleys, had 4 co-channel sectors all

5 directed inside the area to be tested, plus a few more remote co-channel sector during the drive with the highlevel CCI. The comparison has been performed by binning the results using bins centered on the road. A histogram showing the distribution of errors is given in Fig. 3. The estimated standard deviation is about 3.5 db with the average error value of less than + 2 db. The error includes the granularity of the data and their uneven distribution inside bins that include small-scale terrain features (ravines, bridges, etc). Figures 5 and 6 show the same area with maps for two co-channel sectors; comparing the maps one can see that the measurements overlap, allowing to measure more than one sector points in the same area Series1 Figure 5: First of two sectors Figure 3: Histogram for measurement comparison 7.2. Measurement examples An example of the use of the new is shown in Fig. 4. On this map, larger circles represent signal-level measurements from one sector in the network. On top, smaller circles mark the measurements where BSIC values were successfully decoded. One can see that the area covered by the new method is much wider than the area covered by BSIC points. Figure 6: Second of two sectors, same frequency 8. CONCLUSION A new method for accurate measurement of RF signal levels in TDMA networks in presence of CCI has been briefly described, and examples of coverage maps were compared with the old technique. The described method is used in a new RF coverage system developed by DTI. 9. REFERENCES [1] N. Dolder. US Pat. 6,324,382. Figure 4: Areas covered by new and old methods [2] GSM recommendation