At 09:57 PM 3/6/2024, John Ackermann N8UR via time-nuts wrote:

Do you happen to know if it's too late to get in on this? And if
totality is required? (My QTH will only be 99.997%, but work will be
in totality for 3:44. Steel building and no easy window access for an
antenna, though.)

(Sorry to pile on with the questions...) This looks very exciting,
but I'm concerned I might not have any PCs fast enough. Everything
I've read just mentioned recent i5/i7, which is suitably vague. Any
idea if there are benchmarks for guidance as to should work/might
work/won't work?

Looking through the e-waste piles, I see the following:

CPU                          Geekbench 4    5          6
(single/multicore score)
i7-4790 CPU @ 3.60GHz        4802/16033    1073/3967  1377/4374
i5-6500 CPU @ 3.20GHz        4238/12841    910/3187
Xeon E5-1620 v2 @ 3.70GHz    4079/14979    901/3750    765/2822
Xeon E3-1240 V2 @ 3.40GHz    4257/14133    913/3550    773/2707
Xeon X5650  @ 2.67GHz        2618/12433    541/3243    524/2317
i7-2920XM CPU @ 2.50GHz      3734/11556    792/2926    670/2207
i7-5500U CPU @ 2.40GHz        3929/7399      837/1781    1096/2123
i5-4200U CPU @ 1.60GHz        3368/5944      701/1423    914/1781
i3-5005U CPU @ 2.00GHz        2687/5176      538/1222    709/1495
Atom x6425E @ 2.00GHz        2753/6585      687/1673    548/1272
Atom E3950 @ 1.60GHz          1426/4401      311/1129    264/835

How to know if any would be suitable?

Thanks!

--
newell  N5TNL

Hi Scott --

On 3/12/24 20:44, Scott Newell wrote:

There are two or three different activities going on. One uses dedicated
receivers called "GRAPES" while others use the wsprdaemon network I
mentioned.  You can participate no matter how close you are to totality
-- the ionospheric effects extend over a pretty large area.  'm not up
on all the details but the https://hamsci.org site has a bunch of
information on all of this and I'd suggest checking there.

There is one specific project for stations right on the path that are
doing CHU measurements.  That may require some special hardware.

The HamSci annual conference is next weekend and I think but am not
sure it is being streamed via Zoom and/or YouTube.  There will be lots
of stuff about the eclipse experiments there.  Again, the website has
info about that.

[ snippage ]

A lot depends on just what software you're using.  The challenge isn't
as much CPU as being able to handle the USB pipe, which is something
like 2 gigabits for 32 MHz bandwidth.  The fancy GUI based receivers
might take a lot more horsepower, but I've run the RX888 with ka9q-radio
at 64 msps (32 MHz bandwidth) on an Orange Pi 5 and similar SBCs like
the RPi 5 should work.  The RPi 4 is probably just at the edge.

There are many wsprdaemon stations monitoring multiple bands that use
tiny-form-factor PCs with 6th-ish generation i5 mobile processors.  I
have a little Lenovo i5 that was about $125 from a liquidator.  The
RX888 server process by itself uses about 2/3 of one core.  Virtual
receivers that tap into that bandwidth use almost no additional CPU
(though whatever is consuming the data might).  And, with multicasting
you can have receiver processes on other computers on the LAN.

Did I mention that it's cool? :-)

John

hello Scott,

I have an rx888 in my posession for quite a while now. Initially I used it with an older HP laptop, i5 @2.3 GHz. That worked, but only if I stayed at 16 MHz wide sampling. Going wider bandwidth made for lost samples.

Meanwhile my shack has an HP i7 @ 3 GHz, a hexacore I think. That works full bandwidth without skips.

I found that having a fast USB3 interface (as in a sensible datapath on the  mainboard, not just a blue USB connector)more important.

So, I would stay away from the Atom boards, the i3 and probably the Xeons (are those server mainboards). Pick the i7 boards and try.

73 Wilko
PA1WBU

At 08:42 PM 3/12/2024, John Ackermann N8UR wrote:

I did read through a bunch of the hamsci google groups messages and
am considering a GRAPES1 (from the emails, it sounds like they're
really cutting it close on GRAPES2), but I'd have to rig a GPS locked
LO. The RX888 is appealing due to grabbing all the WWV/CHU frequencies.

I do have an uncommitted RPi4 on hand.

I'm wired gigabit here at the house, so that shouldn't be an issue for me.

Yeah, I think you and Mr. Bulte have convinced me to see if I can
find an RX888 vendor that can deliver in time.

At 02:07 AM 3/13/2024, Wilko Bulte wrote:

I've been there before with a Saleae Logic Pro 16 USB3 streaming
logic analyzer--USB3 card wouldn't keep up, and they tracked it down
to the PCIe slot not being fast enough.

The Atoms are cute little Compulab Fitlet2 and Fitlet3 mini-PCs that
we use for industrial automation projects. The Xeons are
decommissioned CAD boxes (Dell Precision workstations--I still like
'em for daily drivers and fileservers due to the ECC memory). The
i7-5500U is a busted up laptop, but the i7-4790 is a decent desktop
(not sure why it was retired...maybe someone wanted a new PC?).

Thanks for the advice!

--
newell  N5TNL

On Mittwoch, 13. März 2024 01:44:37 CET Scott Newell wrote:

That depends a lot on what you're trying to do…

An easily reachable USB3q port that connects directly into the chipset without
any intervening hub chip of questionable provenance would be my first stop,
don't skimp on memory, 24 GiB would be my minimum spec.

I'd cross off anything before Haswell and that includes all your Xeons I
think.  Haswell is fine if you pair it with either Linux or up to Win10 and
the i7-4790 is the one that has (almost) everything enabled that generation
has to offer, including ECC memory (the mobo has to support it).  There's good
deals for Haswell to be had still since a lot of these will not run Win11,
mostly due to lack of a TPM2.  The Broadwell and Skylakes might be interesting
if they have exactly the formfactor and periphery that you're going to use,
otherwise not much of a step up (if at all) from the 4790.  The Atoms are only
interesting if you are trying to go for 24/7 with the lowest possible power
draw, the Elkhard Lake is of course the better choice especially if you want
to use the graphics.

Regards,
Achim.

+<[Q+ Matrix-12 WAVE#46+305 Neuron microQkb Andromeda XTk Blofeld]>+

SD adaptation for Waldorf rackAttack V1.04R1:
http://Synth.Stromeko.net/Downloads.html#WaldorfSDada

The amount of memory is rather irrelevant for the SDR task at hand. I ran my rx888 initially with 8GB RAM and that is no issue at all.

Not that more RAM hurts, but necessary it is not.

Wilko
PA1WBU

Hello, I recently joined this mailing list but I don't think I have said
anything yet :)

I don't know what kind of SDR you are referring to, but if your goal is
time synchronization (or even some coarse positioning), a cheap RTL-SDR
with a Bias tee is actually capable of receiving GPS(+QZSS), which in
theory is more accurate than LORAN and stations such as WWVB and WWV, since
meter-level precision fixes are on the order of nanosecond-precision.

The cheapest RTL-SDRs with Bias tees (and aren't clones or being resold or
anything) are the Nooelec NESDR SMArTee v2 and the RTL-SDR Blog V3 for the
same price as each other ($33.95), the Nooelec one allegedly has a 0.5 ppm
TCXO, while the RTL-SDR Blog one allegedly has a 1 ppm TCXO, however the
Nooelec one doesn't have direct HF sampling (if I understand their listing
correctly).

As for SMA active GPS patch antennas, there is no shortage of listings
online, however I personally have the "GNSS antenna square for
GPS/GLONASS/Beidou/Galileo" ("AN_GPS_A236") from "Suzhou Maswell
Communication Technology Co. Ltd" ("Maswell") which has
GPS+GLONASS+BeiDou+Galileo L1/E1 (and overlaps with some Iridium stuff if
you want to experiment with that too), allegedly IP67, and appears to be
selling right now for $28.

From the same company, there is also the "GNSS IP67 Waterproof GNSS High
Gain Positioning Antenna" ("AN_GNSS_401") which has the same
constellations, is allegedly UV hardened, and most importantly the coaxial
comes through a M12 screw, so you can drill an M12 hole somewhere (e.g. on
a permanent fixture outdoors), seal it, and have a weather-tight solution.
It was the same price as the other when I bought it, however now it appears
to be selling for $36.

I am not sponsored by "Maswell", I just think these two antennas aren't
ugly like many of the other chinesium ones, and upon examining them
first-hand, they appear to have fine build quality for the price. Also if
you have an SMA dedicated GNSS receiver, you could track all
constellations' L1 band signals using these antennas.

Anyways, that $61.95, and if you swap the "Maswell" antenna for a cheaper
one, $42.95. What can you do with these ~$50? Receive GPS signals!

There is software such as GNSS-SDRLIB (Windows) and GNSS-SDR (Linux) which
can get a GNSS fix using your SDR, GNSS-SDRLIB seems to work better
initially but is dated and has issues, while GNSS-SDR takes more tinkering
and editing configuration files but is very flexible and works really well
once you get it to work.

The final question is, what is the quality of the time fixes with SDR?
Obviously, processing baseband GNSS signals (~2 MHz sample rate, 3.2 if you
have a USB interface with superior jitter that doesn't drop samples) has a
higher load and latency than a narrowband Loran or NIST timecode, but when
I look at the USB traffic on my Linux machine when utilizing my RTL-SDR, it
appears to poll (64 bytes) and then get a signal back, here are the packet
sizes I found at the following sample rates:
3.200 MHz: 15936
2.880 MHz: 14400
2.560 MHz: 12864
2.400 MHz: 11840
2.160 MHz: 10816
2.048 MHz: 10304
1.920 MHz:  9792
1.792 MHz:  9280
1.536 MHz:  7744
1.024 MHz:  5184
0.250 MHz:  1088

It appears it also sends ~375 packets/second no matter what sample rate I
choose, except 0.250 MHz (don't choose that), so like ~2 milliseconds of
buffering? Which is terrible, yes, but if you can find a way to sync a
precision PC clock measurement (such as POSIX clock_gettime) to the start
of the buffer, and then again at the end of the processing chain, and then
retroactively looking to see what the PC clock offset would be at the start
of the buffer, I think it would be feasible to get microsecond or even
nanosecond precision time synchronization. Also, you could run this for a
while against something like all of NIST's NTP servers (which I currently
do because I haven't implemented this, and I don't have a dedicated
GPS-locked NTP server or anything yet, with my PC's 20 ppm clock and Chrony
it keeps it to within ~154.5 us 75% of the time using all of NIST's NTP
servers except the university ones which appear to have inferior jitter,
with my PC power cycling (I put it to sleep at night))

In addition to GPS I have also gotten GPS+Galileo locks using GNSS-SDR,
however since the E1 signal is BOC, ~4 MHz receiver bandwidth minimum is
required to practically use it (that is, for it to help the quality rather
than hurt it).
Theoretically L1 QZSS should be identical to L1 GPS, but I live in Idaho
and they never get above 5 degree elevation for me.

This has been my two cents, I didn't go nearly as much into depth about
this as I possibly can, I hope to send another e-mail about my findings
regarding time synchronization using cheap SDR hardware, and possible
investigations into the method I outlined above.

On Wed, Mar 6, 2024 at 7:10 AM JOHN HARTZELL via time-nuts <
time-nuts@lists.febo.com> wrote:

At 02:07 AM 3/13/2024, Wilko Bulte wrote:

Good advice, it turns out!

The RX888 arrived Monday, but I didn't get to play with it until
Tuesday. 64 MHz sample rate on the Xeon E5-1620 brought it to its
knees. The i7-4790 seems ok at a 64 MHz sample rate (load avg 0.30,
top shows 46% CPU) with a locally generated fftw wisdom file (but I
do wonder if I should re-run it for 2 or 4 threads). Then again, I
could be dropping samples left and right and I'm not sure I'd even know...

I ran a 50' wire from the window to the corner fencepost and now that
the sun is down I'm getting decent reception on the WWV and CHU time
stations. The ka9q-web app is a lot of fun!

Now I'm wondering how to GPS lock it. I understand there's a
connector (U.FL?) on board for an external 27 MHz reference. If the
Si5351 clocks the ADC and the software can program it, maybe it can
be tricked into using a 10 MHz reference instead?

--
newell  N5TNL