Data processing: measuring – calibrating – or testing – Calibration or correction system
Reexamination Certificate
2002-02-15
2004-03-23
Shah, Kamini (Department: 2857)
Data processing: measuring, calibrating, or testing
Calibration or correction system
C341S144000
Reexamination Certificate
active
06711515
ABSTRACT:
FIELD OF THE INVENTION
The invention relates generally to the smoothing of results measured as a function of time. Particularly the invention relates to the smoothing of measurement results with the aid of such a measure of the center of a sample which is not sensitive to individual deviating measurement results.
BACKGROUND OF THE INVENTION
The network elements of wireless mobile networks measure at regular intervals for instance the strength, quality and interference level of the radio signal. Changes in the state of the network are then made on the basis of the measurement results, if required. For instance, a decision to shift a wireless terminal's connection from one base station to another (a handover), or a decision to control the signal's transmission power is made on the basis of such measurement results. Thus measurements made at regular intervals are an essential and necessary part in the operation of the radio access network.
The measurement results generally follow a certain, often unknown probability distribution. Thus the individual measurement results vary slightly, also in a situation where the quantity to be measured remains constant and when there are no extra interfering factors. The aim is to estimate the real value of the measured quantity on the basis of the measurement results. Thus the situation is similar as that in estimation, where one on the basis of a sample having a certain size tries to define parameters which describe the probability distribution behind a phenomenon. For instance, the sample mean is an estimate for the average of the distribution behind the measurement results.
There may also occur large variations in the individual measurement results also as a result of for instance occasional disturbances. Thus individual results measured as a function of time are generally smoothed by averaging. The averaging is made so that small successive sliding samples are taken from the measurement results, and the average is calculated from these samples. Thus the individual measurement results are replaced by the average calculated with the aid of a few successive results, and so the effect of at least a part of the individual deviating values (outliers) can be eliminated. If the measurement results are not smoothed, unnecessary changes may be made in the network status on the basis of these individual deviating measurement results.
The prior art network elements of the radio access network use the average to smooth the measurement results: a certain amount of successive measurement results form a sample from which the arithmetic average is calculated. The measurement results are not processed before the calculation of the average. Deviating values can in principle be omitted from the measurement results, but this requires extra logic, in order to decide which measurement result values deviate sufficiently and can therefore be omitted, and extra calculation in the processing of the measurement results. Different measurements and smoothing of the measurement results are made so much in the network elements that the processing of the measurement results must be as straightforward as possible.
FIG. 1
shows a prior art smoothing of measurement results using a moving average. The horizontal axis in the figure represents the sequence number of an individual measurement and the vertical axis represents the measurement result. The individual measurement results shown by dark circles are generated by simulation. In radio access networks the measurement results are often presented as integers between 0 and 63, i.e. as binary numbers with a length of six bits. The individual measurements shown in
FIG. 1
are generated as follows: first there is generated raw data with the aid of a normal distribution having the average 30.5 and a standard deviation of 7, and then the raw data is averaged with an arithmetic average using a sample size of 4. This provides as a result a measurement point set which has a normal distribution with the average 30.5 and the standard deviation 3.5. For instance in a network element of the radio access network the radio signal strength measurement could produce a measurement result corresponding to the measurement point set shown in FIG.
1
.
The measurement results shown by dark circles in
FIG. 1
are smoothed using a moving arithmetic average. As an example the sample size used in the smoothing is 4. In
FIG. 1
the moving average is marked by a uniform line, and it follows the trends of the measurement results with a short delay. This delay is caused by the fact that the four previous measurement points, or more generally the number of measurement points determined by the sample size used in the smoothing, always have an effect on the smoothing result at a certain moment.
A problem in the use of the prior art moving average is that if a sample contains even one value which deviates considerably from the other values, then the sample average will change considerably. The size of the change is further proportional the fact how much the said single measurement result deviates from the other measurement results in the sample. An individual measurement result, which is measured for instance in a disturbance situation, may have a so strong effect on the moving average that for instance the transmission power is interpreted to require a large change. The next moving average which falls within the approved limits will then again cause the transmission power to return to the original value. High requirements are put on the power control algorithm and other control algorithms so that they behave well in such situations.
Prior art problems relating to the use of a moving average are illustrated in FIG.
2
. In the same way as in
FIG. 1
the horizontal axis in
FIG. 2
shows the sequence number of an individual measurement, and the vertical axis shows the measurement result. Interference peaks are added to the simulated measurement results shown in
FIG. 1
, so that the peaks cause a change of either 25 or −25 units to an individual measurement result. This simulated time series of measurement results corresponds to a situation where an interference factor affects the measurement results.
FIG. 2
shows with a uniform line the moving average, which again as an example is calculated using samples of four measurement points, and which is sensitively following the interference peaks. Its value increases clearly when a positive interference peak is included in the sample used for calculating the moving average, and correspondingly the value decreases when the sample includes a negative interference peak. If the horizontal lines at the values 25 and 40 on the vertical axis shown for illustrative purposes would represent for instance limits, which if exceeded would cause the radio signal power control algorithm to start to change the transmission power, then in the situation shown in
FIG. 2
the transmission power control would be unnecessarily started several times.
FIG. 3
shows simulated measurement data having more interference peaks than the situation shown in FIG.
2
. To the measurement point set shown in
FIG. 2
there are added interference peaks which cause a change of either 25 or −25 units to the individual measurement results. Even if we from the measurement results shown in
FIG. 3
still can approximately distinguish a clear trend staying between the values 25 and 40, the moving arithmetic average does not follow this trend at almost any measurement point of the set. For the smoothing of the measurement points we have also in this example used a sample size of four.
If the sample contains outliers the size of the window used for the calculation of the moving average has an effect on the size of the change of the sample average: the larger the sample the smaller effect do the individual outliers have. However, the use of a large sample in the calculation of the moving average can result in that the measurement results are smoothed too much, i.e. that essential changes in the measurement results are detected aft
Lehtinen Joni
Vaara Tomi
Cohen & Pontani, Lieberman & Pavane
Nokia Corporation
Shah Kamini
LandOfFree
Robust and efficient method for smoothing measurement results does not yet have a rating. At this time, there are no reviews or comments for this patent.
If you have personal experience with Robust and efficient method for smoothing measurement results, we encourage you to share that experience with our LandOfFree.com community. Your opinion is very important and Robust and efficient method for smoothing measurement results will most certainly appreciate the feedback.
Profile ID: LFUS-PAI-O-3271348