Accuracy+
Creativity Software (CS) is proposing its Accuracy+ algorithms (Patent Pending), based on predictive matching techniques which are essentially a hybrid “fingerprinting-predictive” algorithm using:
Adaptive propagation models:
- Propagation models – adjusted based on the type of environment (ray tracing, input from network planning tools) and time self-learning
- Antenna models – using standard antenna patterns for macro, pico and omni cells.
Fingerprints adding a local geographic influence coupled with a probabilistic spatial distribution where the “variance” is adjusted according to the environment type
Although CS cannot disclose the details of the algorithms (wholly owned IPR), we can provide a summary of the main concepts and figures.
CS is working on these algorithms in collaboration with the internationally renowned University of Cambridge (ranked number 1 worldwide by the QS World University Rankings® - the most trusted university ranking in the world). The Cambridge University Digital Technology Group has accumulated years of experience in outdoor and indoor positioning.
At the same time, our own team in CS is composed of senior engineers with the highest qualifications (Masters and PhD) in Telecoms and signal processing.
Research work revolves around propagation, antenna modelling, signal processing, deterministic and probabilistic optimisation using Scilab simulation environment.
Accuracy+ and Indian market
The Accuracy+ solution is available “out of the box” from day one, providing enhanced accuracy based on cell tower data provided by the network operator. As more location requests are performed, the algorithms self-adjust (self-learning) to learn the environment and adjust the underlying model.
Accuracy+ solution is already providing accuracy results higher than the ones provided by Enhanced Cell Id and the accuracy will be enhanced by ingesting the data collected during routine drive test conducted by operators for planning and optimization purposes
Accuracy+ has been tested in different cluster types and full compliance with the government requirements was able to be demonstrated.
The main criteria for the assessment of accuracy are the density of cells and type of environment. The hypothesis underpinning the accuracy figures are conveyed below.
Creativity Software’s criteria for the classification of clutter types:
Note: these figures are applicable to GSM and 3G.
Contributing factors: Number of Picocells & 2G/3G mix evolution over time (3G is more accurate)
Comparison of fingerprinting and predictive matching
Comparison of Fingerprinting and Predictive Matching
Comparison of fingerprinting and predictive matching
Comparison of fingerprinting and predictive matching
Comparison of fingerprinting and predictive matching
Comparison of fingerprinting and predictive matching
Comparison of fingerprinting and predictive matching
For illustration of our approach, below is a quick benchmark of predictive matching compared to the traditional fingerprinting method in urban and suburban/rural environments. This takes into account the same fingerprint density and without the addition of any post-processing techniques.
The benefits of predictive matching compared to traditional fingerprinting can be observed in all cases. The most significant improvement is observed in the results between the fingerprints in the 50-300m accuracy range but also in some instances removing bigger outliers.
Availability of commercial grade maps would enhance further the results in favour of the predictive matching method, and would lead to an even looser data collection grid. (i.e. need for less finger print data capture activity).
Example with a loose grid of 250-300m:
Case Study: Summary of Tests Results - CS meets government requirements
Case Study: Summary of Test Results - CS meets Government Requirements
Government requirements
Summary table
Although the POC is limited and doesn’t leverage all the elements that would be available in a full implementation in terms of map quality, availability of Timing Advance data, access to the network planning tool, etc. we can observe the following:
1- Dense urban areas:
Accuracy+ solution will require limited amounts of data collection in order to meet DOT requirements, but generally less than traditional (standalone) fingerprinting methods, leading to lower Total Cost of Ownership (TCO).
For extremely dense areas, the 2012 requirements will be met with data collection required only in specific areas along the coast.
2013 and 2014 targets can be achieved with a relatively loose grid, most probably directly issued from the regular drive tests and complemented in specific places with additional data collection.
This is in line with our experience in other markets and was entirely expected. Predictive matching is marginally superior to fingerprinting in dense urban environment.
2- Suburban & rural environments:
2012 targets are covered already, using the adaptive prediction algorithms out of the box – with no data collection required.
2013 and 2014 figures can be covered using a loose grid of data points which would typically be directly extracted from the routine drive testing already performed by MNO, and complemented by better maps. Predictive matching is significantly superior to fingerprinting.
3- Remote environments:
Results can be achieved using enhanced cell-id, with no data collection.
Accuracy+
|
