Let me speak again about instant pace

The reason I mentioned the cycling activity is to point out that the pace issue may be inflicted by the faulty algorithm rather than being an inherent issue with the GPS reception. Well, it could be both, but if cycling produces a cleaner pace, with everything else equal, that means that Garmin messes up the pace in a wrong way when in a running activity. 

K2 said:

I think accuracy-wise, it's hard for a watch to beat a phone. You have more space, more hardware, more battery headroom.

That doesn't explain why more basic, smaller Garmin watches produce a better pace. That doesn't explain why the last generation of Apple watch produces an excellent instant pace. The GPS antenna design is a big part of it, but it appears Garmin does something with the pace on Fenix that makes it messed up.

My suspicion, is that the pace on Fenix is optimized to be more stable at walking speeds, and that it puts way too much weight on the accelerometer input and deriving the pace from the cadence and the stride length. There is some sort of adaptive algorithm that clearly makes the pace way too slow in the very beginning (probably because it starts with the default walking stride length). Then after 5-10 minutes it adjusts and becomes more accurate. That is far more easy to observe when running on trails under tree cover because the GPS reception isn't as good.

Again, everything above is based purely on my observations.

This is 7 seconds of data from my latest .FIT file when the pace in the data field on the watch dropped below actual pace. The GPS enhanced_speed is really low. The data records for each seconds are higher but also to low. And the distance does not compare to either speed per second. So there are clearly a glitch somewhere. 

Green to the left is GPS data and green to the right is from the data record for each second.

I think that the enhanced_speed to the right is the data that is shown in the data field in the watch. Converted to km/h or mi/h.

Actual pace in this section was around 2.900 m/s so none of this data is correct.

Judging by that OP data just looks to be better filtered on the phone. That’s a complete guess just based on logic - it’s completely improbable anyone’s pace would vary that rapidly and to such a degree so hard to imagine how the designers let data could go out as it is there. Unless this is a hugely long run so deceptively dense data. Or the device is broken. 
my pace is all over the place but not to that degree. That said I mostly look in connect which perhaps smooths,  I haven’t looked at raw data. 

There's a reason why cyclists using power meters don't use instant power as a metric. It's too variable. A graph of the output would look just like the second image. And that's why they have 3s or 5s averaging.

And that would be the sensible fix for instant pace using a wrist-worn GPS device with arms moving around at the same time as the body moves.

Phone cells are used for very coarse position fix. It only helps the phone know in which neighbourhood it is. It can't give the accuracy of the worst GPS receiver.

However, phones have high power GPS receivers and bigger and better antennas. Newer phones can fix to four GNSS constellations, and other terrestial correction sources. Garmin devices fix to a single or dual GNSS and GNSS chip power consumption is very low. However, antenna structure and shape etc. is more important I think. As I know Fenix'es don't have the accuracy of Forerunner's. They have metal crowns and different antennas. As they have the same GPS chipsets, metal bezel and other structures probably play an important role.

There is already 3-5 second averaging based on how the pace changes over time. 

I don’t actually see this as an accuracy issue. 

it seems reasonable to assume Garmin aren’t properly pre-processing / filtering this data. This processing happens on most devices dealing in analogue and digital data, not just sports.

The fact it works better in another activity seems like clear confirmation. There’s probably a line or two of code that they just need to reinsert to make the data more useful. 

hi @Mirko. you asked "My question: is the phone more accurate due to the use of GPS+phone cells to calculate the position?"

the answer is that it varies by phone model and it also varies by the app you use. as you know some compromises can be made to extend battery life.

regarding other comments. Garmin has devoted considerable effort to getting position and pace correct - simply changing to bike mode won't work, simply wearing the watch upside down, wont work. Like EVERY other company they still have a way to go to produce accurate instant pace for all reception conditions.

one way of looking at the source of errors is to consider those that we can handle and those which are problematic.

For those that we know how to handle then companies may CHOOSE not to not provide the best technical solution eg by making a  design choice that uses a sub-optimal aerial (maybe because that aerial is smaller or cheaper).. Indeed battery life is generally considered to be more important and choices over the battery impact the gnss/gps decisions that get made.

That mostly leaves the multipath errors which are where the path of the signal is 'wrong' for some reason...be that bending in the atmosphere or bouncing off a building. These signal could be perfectly received by the best antenna in the world but it would think the wearer was somewhere else (like your reflection in the mirror).

Atmospheric errors (eg bending occurring because of different air layer temperatures...think MIRAGE) might be able to be sorted out this year as Garmin and others are likely to introduce multi-frequency, multi-band chipsets (the new Sony). indeed garmin already have released this in q4.2020 in a handheld

The bounced signal is problematic and that affects all of us running near buildings. My understanding is that it IS possible to somehow detect the changed 'spin' of a bounced signal and thus that theoretically can be corrected for.

regarding other comments about whether or not we need instant pace or lap pace, then it's best to assume that not everyone is like you and that some people think they NEED it and that others don't care, everyone else falls in the spectrum in between. the ONLY way to get reliably correct (instant) pace for running is from a properly calibrated footpod.

I don't know if Garmin will let me post this but i have tested just about every GPS watch model over a standard, complex route (as well as testing elsewhere): https://the5krunner.com/2016/11/05/test-route-for-gps-devices/. surprisingly the Garmin forerunner 745 appears better than other garmin models, you will also see in the results table that the apple watch 6 appears to be very good, however my strong suspicion is that Apple has the cleverest smoothing algorithm on the market which appears to MOSTLY give good track results but, of course, those post-workout pretty GPS track plots may not reflect the bad instant pace you experienced on the exact same run.