Page 28 - Noisy-User-Manual-2017-en
P. 28

26                                               2 Program concepts

            2.2  Determining          the     percentile      confidence
                   intervals

            2.2.1  Problem
            For detailed measurement of noise level trends, use can be be made not only
            of  the  mean  value  but  also  of  percentile  levels  (statistical  levels)  as
            distribution measurement variables, for example, LAF1 or LAF95. By reason
            of  the  usually  stochastic  fluctuations  of  the  noise  level  trend  and  the
            necessarily finite measurement duration, the measuring certainty of percentile
            levels  is  inevitably  limited.  This  is  also  true  for  all  other  measurement
            variables and all parameters derived from them.
            Appropriately,  the  measurement  uncertainty  is  marked  by  a  two-sided
            confidence  interval,  i.e.,  a  figure  comprising  the  percentile  value;  with  a
            specified probability, e.g., 80% or higher, the true percentile value is within
            this  confidence  interval  [1].  In  this  way,  the  limit  of  the  measurement
            certainty  of  the  percentile  can  be  quantified  directly  out  of  the  stochastic
            curve, i.e., without repeating the measurements [2].
            Due  to  the  necessary  computing  workload,  this  quality  parameter  can  be
            realized  in  practical  measuring  applications  only  by  means  of  a  computer-
            based,  user-friendly  measuring  and  evaluation  process  in  accordance  with
            today's PC standards. To meet this requirement, Wölfel Monitoring Systems
            GmbH  +  Co.  KG  developed  the  NOISY  software.  The  new  measuring
            technology was then submitted to a field test [3].


            2.2.2  Basic algorithm: Percentile confidence intervals
            The upper and lower confidence limit Lq,o and Lq,u of a percentile level Lq
            are determined by means of


            L q, o   L  L  L q, u  t   n 1;1 /2  dL  ˆ    q 2 s 2   q 2  s 2     
                          q
                     q
                                                           w
                                                                 w
                                                                    u
                                                        u
                                               dq
                                                    T 

            i.e., based on data that can be observed and data that have to be specified [2].
            Here, q stands for the exceedance ratio (0<q<1); T stands for the measurement
            time; n stands for the number of the time intervals ui or wi present in T, which
            are assumed to be stochastically independent, in which the noise level falls
            below or exceeds the percentile, with n > 5; dL/dq stands for the increase of
   23   24   25   26   27   28   29   30   31   32   33