[EMAIL PROTECTED] wrote: > Hi Bruce, > > again, I did say in my original post that my math only works if their noise > is Gaussian, and that I did not know if it was. > > Also, please note that they do claim 25ps Accuracy , and 800 femtoseconds > Hardware resolution, not the other way around. > > They claim +-25ps accuracy with 3-sigma confidence. The user manuals are > available on the Wavecrest website. > > There is no mention in their documents that averaging is needed to make use > of their 800 Femtoseconds hardware resolution; quite the contrary in some > older white-paper documents they show how they achieve this single-shot > resolution (when looking inside one of these units it's quite amazing > technology). > > Keep in mind that these instruments are on a totally different level than > the 53132A or SR-620: these units cost anywhere from $50,000 to $90,000 new. > > BTW: the only point I was trying to make is that you sometimes can get these > things for <$500 and they are a lot better than the above mentioned > instruments for time-interval measurements. > > Bye, > Said > Said
The simplest way of comparing the counter specifications is to compare their rms measurment noise with say a +7dBm 10MHz input signal (eg from an HP10811). Wavecrest is likely to have a trigger jitter ~ 10ps rms (when the input comparator noise is taken into account with the finite input sinewave signal slew rate) HP53132A has a trigger jitter ~10 ps rms (difficult to work out from the specifications as the input circuitry noise isn't specified however internal jitter (3ps) is comparable to the Wavecrest) SR620 will have a trigger timing jitter ~ 8ps HP5370A/B has a trigger timing jitter ~ 35ps (dominated by the internal hardware jitter trigger jitter is about 3.3ps rms with this input signal. WAVECREST The 800fs hardware resolution is largely irrelevant when the input circuit produces a jitter of that is 12x larger than this. As long as the resolution is better than the rms timing jitter it will not have a significant effect on the measurement noise (~ 10ps rms). 53132A Since the hardware resolution of the 53132A (~ 150ps) is much larger than its timing jitter (10ps) with this signal, the resolution will dominate the measurement noise (~45ps rms). Its a pity that the resolution is so poor otherwise the performance could have been comparable to the Wavecrest. SR620 Since the hardware resolution of the SR620 (~ 25ps typical ) is comparable with its trigger timing jitter (8ps) with this signal, both will have a significant contribution to the measurement noise (~11ps rms). Its a pity that the resolution is so poor otherwise the performance could have been comparable to the Wavecrest. 5370A/B The 20ps hardware resolution has little effect on the measurement noise (35ps typical). Thus for this signal the Wavecrest is perhaps only about 4.5x better than the worst of the other counters, not >100x as you claimed earlier. With higher frequency and or higher slew rate input signals the Wavecrest performance should be somewhat better. Since Wavecrest, 53132A etc have no specifications for the effect of the input circuit noise with a finite slew rate input, the only way to make a more precise comparison is to actually make some measurements. The integral and differential nonlinearity of the Wavecrest do not seem to be specified, nor are the channel delay mismatches. Are thes internally calibrated? Counter internal delay channel mismatches and other systematic errors (as in the HP5370A/B and the SR620 etc) can be measured and corrected. However such constant channel delay mismatches have little or no effect when measuring the Allan variance of a source, nor do integral and differential nonlinearity over portions of a counter's range that are not used in such measurements. If one were to design/build ones own time interval counter, then a resolution of 10ps is possible using a CMOS TDC, whereas subpicosecond resolution should be possible using dual ADC's sampling a pair of ~200MHz sine waves in quadrature phase. Comparator jitter of around 100-200 fsec is possible with a input signal with a sufficiently large slew rate. NB it has been assumed that the input signal noise is negligible (< 50uV rms). The effect of timebase reference noise has also been neglected. This should be negligible if a 1 period (of the input signal) time interval measurement is made. Bruce _______________________________________________ time-nuts mailing list [email protected] https://www.febo.com/cgi-bin/mailman/listinfo/time-nuts
