Bonjour,
je suis confronté à la compréhension d'un système d'intervalle de confiance (binomiale ou Chi² je ne sais pas trop). J'ai déjà travaillé avec des intervalles de type gaussien. Là c'est un peu différent. Il s'agit d'un intervalle de confiance d'un taux d'erreur de données binaires. Je ne comprends pas ce qui est désigné par BERlow et BERHigh. Il semblerait que ce soit l'intervalle de confiance [BERlow, BERHigh]. Or, il semblerait aussi qu'il y'ait une distinction entre ces deux nombres et le véritable intervalle de confiance [lower confidence level, upper confidence level]
Voici le texte en question:
Confidence BER tests are based on the statistical independence of the bit error
events: The probability of a bit error is equal for each received bit. The exact bit error rate is given by the limit of the ratio ne/ns, where ne denotes the number of bit errors and ns the number of received bits, and where the number of received bits tends to infinity. Any real bit error rate test necessarily provides an approximation to the exact bit error rate.
Given the statistical independence of the bit error events, the probability p for any number of bit errors ne at a given exact BER and a given number of received bits ns can be described by a binomial distribution or an appropriate approximation, such as the Chi Square distribution. Estimates of the exact BER are based on this distribution.
Confidence level:
The distributions p(ne, ns, BER) provide a confidence interval [BERlow, BERhigh] for the exact BER at any ne and ns. The confidence interval has the following meaning:
• The probability for the exact BER to be larger than BERlow is larger than the lower confidence level.
• The probability for the exact BER to be smaller than BERhigh is smaller than the upper confidence level.
In the figure below the lower confidence level is the sum of all probabilities p(ne, ns,BERlow) up to the measured number of bit errors. The upper confidence level is the sum of all probabilities p(ne, ns, BERhigh) above the measured number of bit errors.
Pièce jointe 302769
Example: Assume that 20 bit errors were measured after 1000 received bits (ne = 20, ns = 1000, preliminary BER = ne/ns = 2 %). The probability of the exact BER to be smaller than 1.1 % is 1 % (lower confidence level: 99 %). The probability of the exact BER to be larger than 3.3 % is 1 % (upper confidence level: 99 %).
If the specified upper BER limit is 1 %, then the measurementcan be stopped and the receiver can be failed with a risk of less than 1 % that the exact BER is below the specified limit.
Quant à ce dernier exemple, je ne comprends vraiment pas. Est-ce que le 1.1% et 3.3% sont donnés au hasard ou bien s'agit t-il d'un résultat lié à la distribution binomiale/chi square ?
Quelqu'un peut il m'éclairer ?Merci d'avance !
-----