2

The inter-arrival time of a Poisson Process, $t$, conforms to the exponential distribution, so the probability density function for $t$ is $f(t)=λe^{−λt},~t>0$. ($λ$ is the arrival rate of the Poisson Process.)

Next we aggregate the requests according to the following pattern: from the first arrival, within the fixed-length time interval $T$, the requests in this interval are aggregated to the first arrival. In other words, the requests in this interval are removed except the first one. This procedure repeats for the rest arrivals. The following figure illustrates this aggregation pattern. enter image description here

My question is, what is the distribution of the inter-arrival times after the request aggregation?

Bloodmoon
  • 469

1 Answers1

3

Memorylessness might suggest that if the inter-arrival time in the second process is $X$ then it is simply shifted up by $T$ so has density $$g(x)=\lambda e^{−\lambda(x-T)} ,~x\gt T$$

Henry
  • 157,058
  • Clearly because the situation after the waiting time $T$ is exactly the same as at time $t=0$. – Tom-Tom Jun 30 '15 at 07:28
  • Henry, I can to some extent understand this answer, would you please explain it a little more? – Bloodmoon Jun 30 '15 at 08:06
  • 1
    @Tom-Tom Let $Y$ denote the waiting time immediately after T for the next request, because of the memorylessness, $Y$ has the exactly same distribution of the original inter-arrival distribution of the Poisson process. This is what you mean, right? Now, I can derive the distribution of $X=T+Y$ by $P(X<x)=P(T+Y<x)=P(Y<x-T)=1-e^{-\lambda (x-T)}$. This is what I thought. – Bloodmoon Jun 30 '15 at 08:15
  • What if the original inter-arrival time $t$ conform to a general distribution? I have a follow-up question here: http://math.stackexchange.com/questions/1344277/aggregate-arrivals-from-a-renewal-process Would you please take a look? – Bloodmoon Jun 30 '15 at 08:23
  • @Bloodmoon: Your expressions look reasonable to me. The general distribution question will be much more complicated. – Henry Jun 30 '15 at 17:19
  • @Henry I see. Let's still stick to this question. What if $T$ is a random variable? Assume that it also conforms to the exponential distribution with PDF $f_T=\theta e ^{-\theta t}$, will the distribution for $X$ be: $g(x)=\int_0^\infty \lambda e^{-\lambda(x-t)} \theta e ^{-\theta t},dt$? – Bloodmoon Jun 30 '15 at 17:34
  • @Bloodmoon : I suspect you need $\displaystyle \int_{t=0}^x$ so not to $\infty$ – Henry Jun 30 '15 at 20:26
  • @Henry ahhh, yes! $T$ should be less than $X$: $0<T<X$. – Bloodmoon Jul 01 '15 at 05:53
  • @Henry As for the general distribution, if the accurate result is difficult to derive, is there any approximate ways? Or how about just analyzing some typical distributions, like Erlang distribution with shape parameter = 2? – Bloodmoon Jul 02 '15 at 17:59