Testing GPSDO signal performance?

[elroy]

I'm interested in running comparisons to measure how my GPSDO's perform when the antenna is placed in different locations.

I can do this subjectively by eyeballing roughly how many satellites are seen by the GPSDO's over time, but it would be interesting to measure in a more structured way. Is there a way to use Lady Heather or other software to log signal levels over a given period of time (maybe a few hours) so they can be summed without too much trouble? I thought I'd be using a Trimble Thunderbolt, but also have some other options for the GPSDO.

[Zenwizard]

Lady heather will do this but the data is not stored it is cached. So closing and reopening the application will blank the slate. The heat maps are pretty good over a 24Hour period.

Zen

[MIS42N]

If your GPSDO outputs NMEA data you can use a terminal program like Tera Term or PuTTY to write it to a log file. Some programs that process NMEA data will read a log file (e.g VisualGPS). I don't know of one that does exactly what you want (but I haven't looked).

Don't place too much value on the results. Signals can get reflected, producing reasonable strength but not much good for calculation. On the other hand, people have commented that they notice no reduction in performance when the signal level drops (e.g. inserting a splitter to run several GPS units from one antenna)

[pdenisowski]

Assuming your GPSDO gives you access to NMEA strings, you could look at the "$GxGSA – DOP and Active Satellites" strings.

The various flavors of DOP (dilution of precision) will give you a more accurate predictor of accuracy than just satellite count, but DOP will also be a function of the satellite geometry and some other factors as well.

From a ublox eval kit:

$GNGSA,A,3,78,70,69,84,83,68,85,79,77,,,,1.13,0.61,0.95*19

PDOP = 1.13
HDOP = 0.61
VDOP = 0.95

[MIS42N]

I was using VisualGPS (a legacy program) to look at NMEA data, it summarises some DOP information. I was more interested in the plot function where it plots the changing position reported by the GPS. If it is within a couple of meters, the reception is good. As the reception gets worse, the position wanders 5 or even 10 meters.

Anyhow, I decided to look at DOP data for a day. I extracted the $GPGSA records from an NMEA log using command language

Code: [Select]

C:\Users\Alan.DESKTOP-B2TTK1H\Documents>findstr "$GPGSA" 221004nmea.txt > y.csv
Then loaded the data into a spreadsheet with delimiters comma and asterisk.
Using the default graph tool, easy to plot a day of DOP. I haven't cleaned up the result, this was a quick proof of concept. Just deleted columns A to O leaving the DOP data. Result attached.

[nctnico]

I was using VisualGPS (a legacy program) to look at NMEA data, it summarises some DOP information. I was more interested in the plot function where it plots the changing position reported by the GPS. If it is within a couple of meters, the reception is good. As the reception gets worse, the position wanders 5 or even 10 meters.

I agree. AFAIK a GPSDO should first get a location fix to go into fixed position operation. This means that the position wander should give some indication on reception quality.

[rockwell]

U-center from U-blox does a good job in logging data.
It works also with non U-Blox receivers.
It's free to download at their homepage.

[mendip_discovery]

What is the end goal?

Are you looking for time or frequency accuracy?

[elroy]

What is the end goal?

Are you looking for time or frequency accuracy?

Speaking as the OP, my goal is to make a general comparison of GPSDO performance when its antenna is placed in each of several possible locations, in order to select a final location for the antenna. If it's not too involved, it could be interesting to break out the comparison by specific performance areas (i.e. time vs frequency).

[mendip_discovery]

What I was trying to get at is that if you are after frequency/time accuracy then maybe you need to measure that as well. Seeing lots of satellites compared to a few might not equal better time/frequency results.

See Evaluating the Frequency and Time Uncertainty of GPS Disciplined Oscillators and Clocks which talks about Velocity issues due to coax length, accuracy in location etc.

[MIS42N]

What I was trying to get at is that if you are after frequency/time accuracy then maybe you need to measure that as well. Seeing lots of satellites compared to a few might not equal better time/frequency results.

To measure time or frequency accuracy, you need a better reference than the device being tested. Time nuts spend big money to get their own maser/cesium beam/etc but for mere mortals that's out of reach. Unless you have a method that overcomes this?

Having a well located antenna with entry level GPS modules is important. I get around 10 times better stability (which I use as a proxy for accuracy)  between an indoor antenna and a roof mounted one.

There's a hierarchy of GPS modules. Non timing/timing/dual frequency, with increasing accuracy (and price). Similarly with OCXOs. GPSDOs, like with most things, you get what you pay for. You want 10MHz +- 0.01Hz (1 part per billion) can be done for less than $40. You want better than 1 part in 10^-11 then be prepared to spend upward of $1000. And if your budget is $40 then antenna position is important and worth exploring.

[elroy]

What I was trying to get at is that if you are after frequency/time accuracy then maybe you need to measure that as well. Seeing lots of satellites compared to a few might not equal better time/frequency results.

Ah ok, got it.

I'll be using the same coax cable, just moving the antenna around. No position is perfect -- there are a moderate number of trees at any of them, so it's a matter of determining where the trees are least problematic.

I guess one possible approach would be to reboot the Trimble Thunderbolt each time the antenna is placed in a new location, let it run for 24 hours, and record the approximate ppt value shown in Heather. I'm sure there are probably other factors that may cause the ppt value to vary. But just as a rough indicator to show the best of several antenna positions, perhaps that would be good enough?

[elroy]

Having a well located antenna with entry level GPS modules is important. I get around 10 times better stability (which I use as a proxy for accuracy)  between an indoor antenna and a roof mounted one.

At my site, an indoor antenna isn't an option -- the ones I've tried have been completely unusable for all of a number of different GPSDOs.

There's a hierarchy of GPS modules. Non timing/timing/dual frequency, with increasing accuracy (and price). Similarly with OCXOs. GPSDOs, like with most things, you get what you pay for. You want 10MHz +- 0.01Hz (1 part per billion) can be done for less than $40. You want better than 1 part in 10^-11 then be prepared to spend upward of $1000. And if your budget is $40 then antenna position is important and worth exploring.

One thing I forgot to mention is that this antenna will be feeding several different GPSDOs, all with different accuracy.  In general, though, I'd say that +/- 0.001 Hz (1 part in 10^-10) is what I'm looking for.

[MIS42N]

I guess I should be more explicit. The "indoor" antenna is an active antenna on a 3m lead on a window ledge with a limited sky view. The "outdoor" is a similar antenna on a roof about 6m above ground. I do have an indoor antenna (a NEO-6M with patch antenna) connected to a clock near a large window. It can take quarter of an hour to get a fix and drops out frequently.

I found the best locations with a Samsung Tablet running an app GPS Test. I walked around spending a few seconds at each likely location. The window ledge is interesting, moving the antenna 20cm changes the signal quite a lot.

1 part in 10^-10 is a reasonable target. The problem is how do you know you've achieved it? All PLL GPS systems agree to a point. Because they are phase locked, the long term deviation disappears. So comparing one with another will tell you a phase difference between two (say 10MHz) signals. That doesn't tell you a lot.

[pdenisowski]

I found the best locations with a Samsung Tablet running an app GPS Test. I walked around spending a few seconds at each likely location. The window ledge is interesting, moving the antenna 20cm changes the signal quite a lot.

Assuming no huge changes in the sky view from 20cm away, this is almost certainly a multipath issue.  20cm is almost one wavelength at the GPS L1 frequency (1575.42 MHz), so it's the right order of magnitude in terms of distance.

The GNSS simulation on R&S signal generators include the ability to simulate multipath and obscuration because these are not uncommon in real-world GNSS signal reception.  We have an entire app note on this topic

https://scdn.rohde-schwarz.com/ur/pws/dl_downloads/dl_application/application_notes/1gp101/1GP101_1E_Obscuration_and_Multipath_GNSS_Receiver_Testing.pdf

[elroy]

I found the best locations with a Samsung Tablet running an app GPS Test. I walked around spending a few seconds at each likely location. The window ledge is interesting, moving the antenna 20cm changes the signal quite a lot.

What data reported by that app did you use to assess the performance at each location?

[MIS42N]

Assuming no huge changes in the sky view from 20cm away, this is almost certainly a multipath issue.  20cm is almost one wavelength at the GPS L1 frequency (1575.42 MHz), so it's the right order of magnitude in terms of distance.

The GNSS simulation on R&S signal generators include the ability to simulate multipath and obscuration because these are not uncommon in real-world GNSS signal reception.  We have an entire app note on this topic

https://scdn.rohde-schwarz.com/ur/pws/dl_downloads/dl_application/application_notes/1gp101/1GP101_1E_Obscuration_and_Multipath_GNSS_Receiver_Testing.pdf

I believe you are right. I ran VisualGPS, which gives a sky plot of satellites, and it quite happily plotted almost the whole sky even though the only direct sky view is between two buildings. I assume the other signals are received by reflection. The same program also plots all the Lat/Long positions, they varied by many meters (I can't remember how much, but more than 10 meters).

Using the same program with the roof mounted antenna, the Lat/Long positions varied by a meter or two (using a U-blox NEO-7M). That was reflected in the GPSDO performance.

My GPSDO doesn't use a PLL. It accumulates data for about 8 minutes then makes a change to the control voltage. This effectively removes the jitter from cheap GPS modules. The size of the change is a proxy for the accuracy. I don't think a PLL based design can make any claims of accuracy without a better standard for comparison.

My design philosophy is based on the question 'what is the cheapest GPSDO that provides a usable output?'. So all experiments are done with $10 GPS units, $3 antenna, $4 OCXO etc. I know I can get better results by spending more money, but for what? 10^-10 accuracy can be used for a down converter for 10GHz signals, with maximum 1Hz error. Who needs more?

I am tempted to build a quadrafiliar antenna, as it reputedly rejects reflected signals. But you know ... time.

[MIS42N]

I found the best locations with a Samsung Tablet running an app GPS Test. I walked around spending a few seconds at each likely location. The window ledge is interesting, moving the antenna 20cm changes the signal quite a lot.

What data reported by that app did you use to assess the performance at each location?

The app gives signal strength for each satellite and an estimate of the accuracy of the lat/long. It is crude but eliminates all the less useful antenna positions.

[JOFlaherty]

Putting a GPS antenna on the roof, and cabling the 1.5GHz to a Ublox module could give better reception, but there would be attenuation on the 30ft or so of cable. 
 How well would it work to locate the Ublox module on the roof, and cabling the 1 pps signal to the GPSDO circuitry?

[MIS42N]

Putting a GPS antenna on the roof, and cabling the 1.5GHz to a Ublox module could give better reception, but there would be attenuation on the 30ft or so of cable. 
 How well would it work to locate the Ublox module on the roof, and cabling the 1 pps signal to the GPSDO circuitry?

I have tried this. It works OK. I used a line driver and receiver (UA9638, UA9639), connected by Cat 5 Ethernet cable. The cable is wired with two pairs for signal (NMEA data and 1pps), two pairs for power (12V with a 7805 regulator at the roof end to feed the GPS and UA9638). Signals terminated with 100Ω at the UA9639 end. Tested up to 20 metres. The design of the PCB for my GPSDO includes the traces for the receiver end including the Ethernet connector as an alternate for direct connecting the GPS module. To swap between the two is easy, the UA9639 is in a socket. To run with a local GPS, pull the UA9639 and connect the GPS via a 5 wire pin header. To run with remote, pull the 5 wires, insert the UA9639.

There's a signal delay, but I don't know if it changes (with temperature, voltage, etc.). As long as it is constant or nearly so, it doesn't really matter how long it is. I haven't seen a problem.

I still use an active antenna, it is on the roof. The receiver and associated bits are under the roof.

[JOFlaherty]

There's delay with the RF connection, too. The ethernet type cable sounds like a good idea, with enough more pairs so you can get serial communication and power, too. I think I'll try it.