hi Sharmistha, I think you are getting confused between the actual error between the mote's clocks and the prediction error as used in this paper. Let me take a stab at clearing out your doubts. Let us represent the clocks of two motes by c1(t) and c2(t). Here t represents the real time t, say UTC.
The error between these two clocks at a given time t1 will be given by c1(t1) - c2(t1). This error can be attributed to many factors such as offset in starting the nodes, drift and skew over time. And the worst case / average case numbers (assuming no offset) is something that can be obtained by the datasheets. Now, what this paper does is maintain a relative clock model between these two clocks, some function G(.), i.e. applying G(.) to c1 gives you the expected time in c2. The prediction error, as defined in this paper and in Figure 6 corresponds to c2(t1) - G(c1(t1)). No, FTSP does not have that much amount of error. I think you are referring to the line in Introduction. By sampling at a lower sampling period, FTSP guarantees that the error can never exceed 90 microseconds. The actual error attained by FTSP will be definitely of much smaller magnitude. Please let me know if this is clear. I shall be glad to answer your further queries on the paper. bye, saurabh On Thu, 9 Mar 2006, Sharmistha Maitra wrote: > > Dear All, > > This is a question about the paper 'Estimating Clock Uncertainty for > Efficient Duty-Cyclicng in Sensor Networks', found at > http://nesl.ee.ucla.edu/document/show/153 > > The paper entions - 'time prediction error of a sensor node' increases > monotonically with 'Sampling Period of time synch messages' (figure 6 in > paper). I agree that it will increase monotonically, but how did the > authors arrive at this graph ? I mean how did they arrive at the > particular error values as shown in the graph ? Elsewhere in the paper > they have mentioned that FTSP (a quite popular protocol) with 1 minute > sampling period has prediction error of 90 us. But here in this graph I > see ~5us error for 1 minute sampling period. By what method has this been > calculated ? > > I dont know if this is the right forum to talk about the paper, I > apologise if it is not. Anybody who has read this paper, or the authors > themselves (if they are in the mailing list), if they can clarify , I > will appreciate. > > Thanks, > > Smaitra. > > ----- Original Message ----- > > From: Vinayak Naik <[EMAIL PROTECTED]> > > Date: Thursday, March 9, 2006 12:19 pm > > Subject: Re: [Tinyos-help] Mote Drift > > > A relevant paper. http://nesl.ee.ucla.edu/document/show/153 > > > > - Vinayak > > > > On 3/9/06, Sharmistha Maitra <[EMAIL PROTECTED]> wrote: > > > > > > Dear All, > > > > > > I am working on modelling the clock drift of a mote (mica2 ) > > caused by > > > environmental conditions like temperature. Can anybody guide me > > where can I > > > find any information on this, something like a relation of drift > > amount and > > > operating frequency ? > > > > > > My other question- In general we know that mote frequency can > > drift by 40 > > > ppm (natural drift, not due to temperature)....can anybody give > > me an idea > > > how frequently does this change happen....I mean in the absence > > of all > > > external phenomenon, approx how long will a mote hold on to its > > previous> frequency before 'naturally' jumping off to the next. > > All this because > > > current literatures suggest that short term stability of the > > motes are good. > > > What is the extent of this short term ? > > > > > > Thanks... > > > > > > Smaitra. > > > > > > > > > > > > _______________________________________________ > > > Tinyos-help mailing list > > > [email protected] > > > https://mail.millennium.berkeley.edu/cgi- > > bin/mailman/listinfo/tinyos-help > > > > > > > > > > > > _______________________________________________ Tinyos-help mailing list [email protected] https://mail.millennium.berkeley.edu/cgi-bin/mailman/listinfo/tinyos-help
