Wolf,

Thank you, that is great news (10us pulse compatibility, and 25ns standard 
deviation). I have an E-MU 0202 so I will use the (192kHz centroid 4x 
interpolated windowed-sync-filtered) setup that you mentioned. 

On my previous attempts I was using a Scarlett 2i2 at 96kHz and probably didn’t 
have the 1pps channel configured correctly (in Windows and/or SpecLab); I'll 
provide for some hardware termination in case the Trimble rise time, coax 
length, and Z_in of the sound card don't agree.

I'll also figure out what I did wrong with the Scarlett 2i2; and try the Roland 
SuperUA to see if it has some favorable attributes.

Thanks again,

Jim AA5BW

  

-----Original Message-----
From: [email protected] 
[mailto:[email protected]] On Behalf Of Wolfgang Büscher
Sent: Saturday, October 29, 2016 7:14 PM
To: [email protected]
Subject: Re: LF: Re[2]: LF: Idiot's guide to receiving Ebnaut? - Tutorial part2



Am 27.10.2016 um 21:31 schrieb Paul Nicholson:
>
> I have used the Trimble Thunderbolt with its fixed 10uS pulse.  It 
> worked (vlfrx-tools, M-Audio 192) but timing jitter improved after a 
> pulse stretcher increased the width to 1mS.  As Andy says, the 10uS 
> pulse energy was a bit low giving poor S/N and a jittery centroid of 
> the smoothed pulse.
>
> Now I use a Ublox for timing VLF reception and this works fine.  The 
> Thunderbolt is used for calibration and as a reference for comparing 
> my growing collection of GPSDOs.
>
> I'll be interested to see Wolf's comparison of edge and centroid 
> timing.  I've never had any luck with edge timing.
>

I compared both now, and measured the standard deviation in the GPS sync 
pulse timing.
Used an E-MU 0202 at 192 kSamples/second.
Test result:
  - 100 ms pulses, using *edge* detection (in the fourfold interpolated, 
windowed-sinc filtered signal):
    standard deviation about 40 to 50 ns/second

- 10 us pulses, using *centroid* detection (also with fourfold 
interpolation):
    standard deviation a very respectable 25 ns/second.

The interpolated waveform of what used to be a 10-us-pulse is here:
http://www.qsl.net/dl4yhf/t/10us_pulse_EMU0202_192kS_interpolated.png
  (the green line in the center marks the length of a sample *from the 
soundcard*, the orange segments are interpolated).

BUT...
After reducing the sampling rate from 192 to 48 kHz :
   - standard deviation 200 ns for the edge-detection method

   - standard deviation 600 ns for the centroid method, reasons not 
understood yet
   (this may be a bug, or it may be caused by using integer array 
indices as 'x' coordinate for the centroid detection area.
   I will refine the algorithm for these 'very short' pulses by using 
polynominal interpolation for the centroid edges as well.
   That would be a fairer comparison.)

Even with the above restrictions, 10 us sync pulses seem to be ok, if 
the soundcard uses a good delta-sigmal ADC (which results an almost 
ideal low-pass filter and thus the text-book pulse response as in the 
screenshot linked above).

So, short answer for Jim:
You can use the Thunderbolt E's 10-microsecond sync pulses as they are, 
no need to shape or stretch them.
I will upload a new version of SL soon, but first I will try to improve 
the pulse timing for lower sampling rates.

Cheers,
   Wolf .
(back in "MEZ" / CET now - farewall, nice daylight-saving time.. sigh.. !)