Tracking issue: Update RF acceptance curves

[Vagabond]

This is a meta issue for some cross cutting work to improve our RF acceptance criteria for PoC witnesses.

First, some graphs:

This is the current Free Space Path Loss (FSPL) acceptance curve (in red) against a scatter plot of distance (in meters) vs RSSI in the US915 region. As you can see it is extremely lenient. In yellow you see the curve adjusted for actual output power and finally in green you see the yellow curve with the RSSI values multipled by 1.3 to attempt to account for some real world losses.

Beyond this, we currently use the same curve in EU868 (and other regions), where it's even less applicable:

Clearly correcting for transmit power and frequency are important. We also need to come up with a reasonable loss factor, whether it's simply multiply by 1.3 or something better (arguably 1.4 might be a better fit).

Next, consider the RSSI acceptance curve. Again the current curve is in red. @JayKickliter has made direct observations of the radio by cabling the transmitter to the receiver and adding attenuation. These are the green and yellow lines. One thing of note is that SX1302 reports both channel RSSI as well as signal RSSI. The channel RSSI seems to bottom out at -117 but the signal RSSI can go lower. The miner should be corrected to report signal RSSI (RSSIS) when available.

Another thing to note is that the current curve (which was computed against different data) has a spot where it significantly goes above the observed minimum values. This is contributing to cases where we are improperly rejecting witness reports.

Again, here's the same thing in EU868, note that the current curve has almost no impact because it was computed for US915 and with different transmit options:

Also notice the values seemingly following the RSSIS curve at the bottom, we need to investigate where these are coming since we do not report RSSIS and we believe RSSI bottoms out at -117.

Here's an initial list of things we should do:

• [x] Report RSSIS instead of RSSI when available.
• [ ] Determine loss factor for FSPL (and make it a chain variable). Potentially different loss factors for different regions.
• [ ] Compute FSPL per region using region specific TX power and Frequency.
• [ ] Try to curve-fit the observed RSSIS curves and replace the current acceptance curve with a per-region one.
• [ ] Investigate lower than expected possible RSSI reports in EU868 (which hotspots are reporting these values and why).
• [ ] Allow hotspots to declare their antenna configuration (tx/rx gain) and thread those declared values into the FSPL calculation. Tx power would be lowered for high-tx-gain antennas to preserve regulatory compliance.
• [ ] Build a geographic location to regional parameter lookup table. Right now the miner uses the API to determine this, but we need to have a proper lookup table. Should return a region code eg. US915 that we can then look up for parameters.
• [ ] Characterize other regions we support or plan to support soon. If we change regional parameters (eg the stronger EU868 poc channel proposal) characterize those as well.

[ke6jjj]

Can you label the axes in all the graphs? I'm having a hard time understanding what's being plotted in the first two graphs, in particular.

[Carniverous19]

The red FSPL trace does not seem to be labeled correctly I believe its referring to the formula on line 265 of blockchain_utils that uses 6dB for tx/rx antenna. Note I do not believe this is the formula used for PoCv10 validitiy calculation The FSPL formula(s) on line 270+ are the ones that are used that roughly match the orange trace (I believe they are referenced to TRANSMIT_POWER in blockchain_utils which is still 28dB. You may have adjusted that down to 27 for the proposal). The important point is we are already basically at the orange curve +/- 1dB.

Another suggestion is to repeat the RSSI plots zoomed in to 0-2km ish. A lot of the ability to discriminate (and possibility for flagging honest hotspots) is in this range. That's also where most of the PoC activity is so I would make sure the proposals are accurate within the ranges of most common PoC (also FSPL is pretty poor at differentiating at long distances anyway, its basically a flat line).

Lastly think we should discuss the confusion matrix for these filtering methods (or realistically an empirical guess at what the results would be). Sure we can draw a boundary at a 90th or 95th or 99.99th percentile but is that actually a good boundary for valid / invalid? Are we actually catching gamers or people lying about their location/coverage? Are observations below the proposed orange SNR trace somehow misrepresenting their coverage or are they just a standard deviation further away from an expected value? Similarly, are those with RSSIs above the green trace but below the orange trace gamers or over-achievers?

Last-finally :) I'll echo @ke6jjj suggestion on discord to plot a heatmap vs scatter plot to better understand the distribution of observations. An imperfect but easier way to do this is to add a high transparency to the scatter plot markers. Not truly representative of density but at least will give a rough saturation gradient with dense vs sparse areas.

[Vagabond]

The red FSPL trace does not seem to be labeled correctly I believe its referring to the formula on line 265 of blockchain_utils that uses 6dB for tx/rx antenna. Note I do not believe this is the formula used for PoCv10 validitiy calculation The FSPL formula(s) on line 270+ are the ones that are used that roughly match the orange trace (I believe they are referenced to TRANSMIT_POWER in blockchain_utils which is still 28dB. You may have adjusted that down to 27 for the proposal). The important point is we are already basically at the orange curve +/- 1dB.

The orange line is the red line with the tx power lowered by 1db. I did not use the old 6db function.

Another suggestion is to repeat the RSSI plots zoomed in to 0-2km ish. A lot of the ability to discriminate (and possibility for flagging honest hotspots) is in this range. That's also where most of the PoC activity is so I would make sure the proposals are accurate within the ranges of most common PoC (also FSPL is pretty poor at differentiating at long distances anyway, its basically a flat line).

Yeah, I'll try that. I was using zoomed in graphs for my local exploration but I posted full ones for clarity. I agree that FSPL is pretty bad at long hops, we need something else there (we know that things like earth curvature, dust, air, etc preclude extremely long hops).

Lastly think we should discuss the confusion matrix for these filtering methods (or realistically an empirical guess at what the results would be). Sure we can draw a boundary at a 90th or 95th or 99.99th percentile but is that actually a good boundary for valid / invalid? Are we actually catching gamers or people lying about their location/coverage? Are observations below the proposed orange SNR trace somehow misrepresenting their coverage or are they just a standard deviation further away from an expected value? Similarly, are those with RSSIs above the green trace but below the orange trace gamers or over-achievers?

I actually don't know what's going on with those reports below the orange line, I was going to look up who is reporting those values and see if I can draw any conclusions there.

Last-finally :) I'll echo @ke6jjj suggestion on discord to plot a heatmap vs scatter plot to better understand the distribution of observations. An imperfect but easier way to do this is to add a high transparency to the scatter plot markers. Not truly representative of density but at least will give a rough saturation gradient with dense vs sparse areas.

Yeah, I'll try this as well, good idea,

[Vagabond]

New graphs, with titles, transparency and axis labels:

[Vagabond]

Actually the loss factor was 1.4 on these, honestly 1.5 looks even better of a fit

[Vagabond]

[Carniverous19]

The orange line is the red line with the tx power lowered by 1db. I did not use the old 6db function.

It cant be the traces are not 1dB apart. the FSPL is the same (and independent of tx power) so assuming same tx/rx antenna gains the RSSI at 28dBm tx will be exactly 1dB higher than RSSI at 27dBm tx power. Ill plot the various formulas described in the legend with formulas included for comparison.

Can you point to the code or function used to generate the data for the curves?

What is actually being plotted here is closer to a link budget.

You can see Ptx is just added to the other terms so if Ptx goes up or down by 1dB Prx goes up or down by exactly the same amount. Not the case between the red and yellow curves which are off by ~8dB

[Carniverous19]

My plots with similar axis scale. Note the red line is above -80 dBm at 14km at 6dBi antenna gain and the two traces at 1.8dBi are below -80dBm. Your red trace is definitely above -80dBm (and follows my red trace pretty closely).

Same plot zoomed in around the actual curves. Notice the 27dB Ptx and 28dB Ptx are 1dB apart at every distance. Much closer together than your red and orange traces.

Here is the python code that generated the data for the plots Lfs matches the FSPL formula in line 270+ of blockchain core, I pulled out the antenna gains as sperate terms (edit: changed Prx variable name).

dists_km = np.arange(0.1, 14.1, 0.1)
Lfs = 10 * np.log10(np.power((4*np.pi * 915e6 * dists_km*1e3)/(299792458), 2))

Ptx, Grx, Gtx = 28, 6, 6
Prx = Ptx + Gtx - 0 - Lfs - 0 + Grx - 0
plt.plot(Prx, dists_km, color='red', label=f'Ptx={Ptx} Gtx/rx={Gtx}/{Grx}')

Ptx, Grx, Gtx = 28, 1.8, 1.8
Prx = Ptx + Gtx - 0 - Lfs - 0 + Grx - 0
plt.plot(Prx, dists_km, color='orange', label=f'Ptx={Ptx} Gtx/rx={Gtx}/{Grx}')

Ptx, Grx, Gtx = 27, 1.8, 1.8
Prx = Ptx + Gtx - 0 - Lfs - 0 + Grx - 0
plt.plot(Prx, dists_km, color='teal', label=f'Ptx={Ptx} Gtx/rx={Gtx}/{Grx}')

Formula from line 279 in blockchain_utils for comparison:

10*math:log10(math:pow((4*math:pi()*(Frequency*1000000)*(Distance*1000))/(299792458), 2))-1.8-1.8.

[Vagabond]

I suppose I may have made a mistake with the red curve, which is good. Certainly my point remains that a raw FSPL curve is not really great for this.

[Carniverous19]

cool... so yea the orange or my blue curve are optimistic and representative of "ideal" values not real world but I think we need more logic in deciding hotspot interactions should not be rewarded than "most people are in this range, you are above therefore you earn 0" There are many legitimate reasons why witnesses may be closer to ideal than 95% of hotspots. 3 sigma away from the mean isnt anomalous if it happens in <1% of occurrences.

Maybe do something like limit = min(ideal FSPL, best fit curve + 6dB margin) in addition to accounting for actual antenna gains (maybe bound by regulatory limits). There needs to be some understanding of why measurements fall above some proposed limit but below ideal FSPL and if those should not be rewarded. Again having better signal than 99% of people != cheating. It may mean top 1% of effort where most people stick them on an end-table behind a screen window on their first floor.

[Vagabond]

Yes, definitely, the FSPL is just an upper bound, there's some kind of distribution after that and you're right that some situations will be better than others. Allowing hotspots to declare their antenna gain will help even the playing field a bit more but we do need some kind of reasonable loss factor to account for losses due to air/vegetation/buildings/etc as FSPL, IIRC, assumes a vacuum without any obstacles.

Things higher than the FSPL predicts are in theory possible because of things like constructive interference, but I think we should rule them out simply because those will be exceptional circumstances.

[ke6jjj]

I concur on one important point: Please do not start enforcing strict FSPL limits unless the ability to report antenna gain is also deployed at the same time. I was going to write up the HIP for antenna gain reporting, but it looks like I won't have to now!

[Carniverous19]

The more I look at the SNR curve the more I think we should just drop the constraint. I really don't think any gamer or out-of-place hotspot will be detected from this curve. Maybe 1 sample in EU at -120 RSSI and +12 SNR, but is that actually a gamer or bit flip somewhere, or just 1:10,000 "lucky" low noise reading.

FSPL + tweaks I definitely see value in and we can see clear gamers in the US but SNR I dont think adds any actual value at least at this level of filtering (maybe for SOM or other more advanced anti-gaming methods so the data should still be collected).

[ke6jjj]

@Carniverous19 "The more I look at the SNR curve the more I think we should just drop the constraint."

I vehemently agree and have been silently hoping for this since the beginning.

[Vagabond]

I'm mostly interested in looking at the outliers to see what's going on there, but I'm inclined to agree.

[nickdifranco]

Any progress on this? What can I do to help change the constraints on RF acceptance?

[kellybyrd]

I've got a hotspot with an 8dBi omni antenna on a chimney and I'm seeing it being called alternately a valid then invalid witness of a few different hotspots around me, the closest is 5km. Is this useful info?

[nickdifranco]

I've got a hotspot with an 8dBi omni antenna on a chimney and I'm seeing it being called alternately a valid then invalid witness of a few different hotspots around me, the closest is 5km. Is this useful info?

Talked with a member of the Helium team on discord. They are working on a fix to the current RF curve that should help those of us with the invalid witness problem. No word on a release but they are definitely aware.

[amirhaleem]

Talked with a member of the Helium team on discord. They are working on a fix to the current RF curve that should help those of us with the invalid witness problem. No word on a release but they are definitely aware.

yup, we're still working on it, just been a little slow due to work on validators and other high priority issues. we're also looking to implement different curves per region - due to differing characteristics like spreading factor, power output, and frequency - which adds a little more complexity.

[GP-Colorado]

Regarding inclusion of an antenna gain value for each hotspot: would there be any reason for hotspot operators to declare anything other than a 9 dBi antenna, regardless of what they are using, apart from personal integrity? 27dBm + 9dBi = 36dB, the maximum allowed in US.

[kwatkins99]

I am a recent hotspot owner and like many others are trying to maximize coverage and earnings. I added a rooftop 5.8dbi antenna and now I am invalid most of the time to my 3 closest hotspots due to hlgh SNR. I read that SNR may be removed as a criteria. I went to a considerable effort to improve my coverage but feel I would be better off going to stock inside. Is there any update on implementation of changes to the criteria?

[kwatkins99]

There doesn't seem to be anything happening with the update. Is it going to happen?

[vihu]

There doesn't seem to be anything happening with the update. Is it going to happen?

There is some work-in-progress proto#77 and core#776. Since the RF acceptance curves would change depending on the region a Hotspot is located in; we've been trying to add blockchain level support for region specific Lora parameters and proceed from there. Don't have a set date yet but this issue has not been abandoned.

[kwatkins99]

I don't know if I can reply, but I want to thank you for letting me know things are proceeding. I downgraded my rooftop antenna to 3dbi from 5.8dbi and still my 2 closest come up invalid most of the time. Since the change doesn't look imminent, I am just going to reduce my coverage and move my OG hotspot inside

On Wed, May 12, 2021 at 1:39 PM Rahul Garg @.***> wrote:

There doesn't seem to be anything happening with the update. Is it going to happen?

There is some work-in-progress proto#77 https://github.com/helium/proto/pull/77 and core#776 https://github.com/helium/blockchain-core/pull/776. Since the RF acceptance curves would change depending on the region a Hotspot is located in; we've been trying to add blockchain level support for region specific Lora parameters and proceed from there. Don't have a set date yet but this issue has not been abandoned.

— You are receiving this because you commented. Reply to this email directly, view it on GitHub https://github.com/helium/miner/issues/618#issuecomment-840011741, or unsubscribe https://github.com/notifications/unsubscribe-auth/ATSX25DAQSZZVZFVECRPBLLTNLDONANCNFSM4XHZ3RUA .

--

J. Kirk Watkins, MBA

Manager

@.***

(314) 852-7494

[mistakeNot7]

One thing I haven't seen mentioned anywhere in modelling the "real world loss" factor against FSL are the major clutter (land coverage) classes: open, rural, urban. Each will get a different factor for modelling FSL to a more realistic figure.

[rbrtio]

Hello all, I am also running into the invalid witness issue due to higher than expected SNR values. I moved from the stock antenna and have tried both a 5dBi and an 8dBi both outside and in my attic with similar results which is frustrating as you might imagine.

I am US based, so operating in the 915Mhz band.

If you guys need any data to help make a determination or if you need more details on my scenario, please let me know.

[jhunter-007]

Am I correct that updating the correct elevation and antenna gain in hotspot assert is not currently being factored into the PoC transaction for valid/in-valid witness? We are seeing several PoC events with invalid witness, even though the RSSI and SNR results are correctly located the valid witness graph we've seen. Makes me think that the current model does not factor in elevation and gain correctly.

[vihu]

Am I correct that updating the correct elevation and antenna gain in hotspot assert is not currently being factored into the PoC transaction for valid/in-valid witness? We are seeing several PoC events with invalid witness, even though the RSSI and SNR results are correctly located the valid witness graph we've seen. Makes me think that the current model does not factor in elevation and gain correctly.

You are indeed correct, currently deployed code only added support to add elevation/gain data to the ledger but does not use it for POC validation; but note that storing that data also does not affect POC validation either. The team will do the release cycles + related chain variable updates and communicate when the code is ready to go out. WIP can be seen in blockchain-core#776.

[jhunter-007]

Am I correct that updating the correct elevation and antenna gain in hotspot assert is not currently being factored into the PoC transaction for valid/in-valid witness? We are seeing several PoC events with invalid witness, even though the RSSI and SNR results are correctly located the valid witness graph we've seen. Makes me think that the current model does not factor in elevation and gain correctly.

You are indeed correct, currently deployed code only added support to add elevation/gain data to the ledger but does not use it for POC validation; but note that storing that data also does not affect POC validation either. The team will do the release cycles + related chain variable updates and communicate when the code is ready to go out. WIP can be seen in blockchain-core#776.

Thank you. I will read more and follow this work. I work in wireless network deployment, as do some of my colleagues. So happy to contribute where it might be needed. (We currently have a few hotspots using 14dbi antennas at high elevations (tower and mountains around metro areas, that are being returned with invalid witness (RSSI too high) at certain distances. So it became obvious that the formula used in PoC was assuming our location assert was not correct and/or a lower gain antenna.). We have run RF simulations and also other field tests proving that in fact, 90-100km range is possible in 915mhz.

[PipArmy]

Invalid witness — (Witness RSSI below lower bound) RSSI SNR Distance -107dBm 10.00dB 1.5 km

[JonPike1]

Interesting stuff, as a RF guy and ham, I do follow the issue. Though I'm puzzled still about the basic reasoning that brought this into being.

Carneverous19: There needs to be some understanding of why measurements fall above some proposed limit but below ideal FSPL and if those should not be rewarded. Again having better signal than 99% of people != cheating. It may mean top 1% of effort where most people stick them on an end-table behind a screen window on their first floor.

I agree, and have concerns, since I'm newly hosting a site the past week or two. Started with simple, lower setups but I have two better spots. A pole on my roof (for ham antenna) that is about 42' above ground. Great spot, maybe? Also have old TV mast about 28' AGL. Currently I have the RAK about 7' above my roof, (about 25' AGL) just using the RAK low gain folding antenna sticking thru a aluminum foil covered box for a temporary setup. (heat shielding and ground plane too!)

Even with that setup, I'm seeing >60 witnesses, stations as far as 50-140km, and yes, usually 2 or so invalid witnesses on close in sites. I worry that I go to all the hassle of getting the RAK and 5.8dbi antenna up on that high pole, and I'll lose a lot of my local stations, because I'm now a "top 1%" and results will be too good. Will I gain back overall what I lose, with greater witnessing? Who knows? I know it will be a fair bit of work and materiel costs at risk to find out. I probably won't do that for now.

Will be watching for any updates, but I agree that this current method is not doing what it should.

[curiousfokker]

Hi Jon, don't worry about it too much. You don't get punished once you get into the top 1%. I think para1 was saying that just because you're in the top 1% doesn't mean you're a cheater. We had serious problems early on with cheaters/gamers. It's almost always worth it to optimize your setup, although keep in mind that location is more important (by far) than elevation or antenna.

[rbrtio]

Interesting stuff, as a RF guy and ham, I do follow the issue. Though I'm puzzled still about the basic reasoning that brought this into being.

Carneverous19: There needs to be some understanding of why measurements fall above some proposed limit but below ideal FSPL and if those should not be rewarded. Again having better signal than 99% of people != cheating. It may mean top 1% of effort where most people stick them on an end-table behind a screen window on their first floor.

I agree, and have concerns, since I'm newly hosting a site the past week or two. Started with simple, lower setups but I have two better spots. A pole on my roof (for ham antenna) that is about 42' above ground. Great spot, maybe? Also have old TV mast about 28' AGL. Currently I have the RAK about 7' above my roof, (about 25' AGL) just using the RAK low gain folding antenna sticking thru a aluminum foil covered box for a temporary setup. (heat shielding and ground plane too!)

Even with that setup, I'm seeing >60 witnesses, stations as far as 50-140km, and yes, usually 2 or so invalid witnesses on close in sites. I worry that I go to all the hassle of getting the RAK and 5.8dbi antenna up on that high pole, and I'll lose a lot of my local stations, because I'm now a "top 1%" and results will be too good. Will I gain back overall what I lose, with greater witnessing? Who knows? I know it will be a fair bit of work and materiel costs at risk to find out. I probably won't do that for now.

Will be watching for any updates, but I agree that this current method is not doing what it should.

Jon,

The problem should get much better soon. There is work in progress to modify the PoC algo that should make the issue much better. The discussion I have seen has a goal of deploying this change near the end of the month, but network stability will always take priority over these types of changes.

[jhunter-007]

I wouldn’t call being in the top 1% cheating - it means careful and efficient placement with quality materials. As long as it’s not over 36 eirp, then it’s ok. The real problem is that if you have a high gain antenna, 100 ft AGL, and a good high elevation, we can get hotspots at 50-60 + miles. The issue is that the algorithm in the Helium PoC still doesn’t factor in higher gain antenna or higher elevation/tower placement. So we see invalid witnesses as we are still reaching 50 miles with -115 RSSI. The Helium PoC formula needs to be adjusted to factor in real RF scenarios, which include higher gain antennas and higher AGL placement.

Get Outlook for iOShttps://aka.ms/o0ukef

---

From: curiousfokker @.> Sent: Tuesday, July 13, 2021 3:36:29 PM To: helium/miner @.> Cc: Jeff Hunter @.>; Comment @.> Subject: Re: [helium/miner] Tracking issue: Update RF acceptance curves (#618)

Hi Jon, don't worry about it too much. You don't get punished once you get into the top 1%. I think para1 was saying that just because you're in the top 1% doesn't mean you're a cheater. We had serious problems early on with cheaters/gamers. It's almost always worth it to optimize your setup, although keep in mind that location is more important (by far) than elevation or antenna.

— You are receiving this because you commented. Reply to this email directly, view it on GitHubhttps://github.com/helium/miner/issues/618#issuecomment-879347876, or unsubscribehttps://github.com/notifications/unsubscribe-auth/AUMNYLDK3SXXOAODISQK2GLTXSIT3ANCNFSM4XHZ3RUA.

-------------------------------------------------------- This e-mail, and any attachments, are a private communication and contain confidential information intended solely for the use of the person(s) and/or entity to which this e-mail is addressed. If the intended recipient is a client, this communication may also be subject to an internationally enforceable non-disclosure agreement. If you are not the intended recipient, or an agent of the intended recipient, any copying or distribution of this communication is strictly prohibited by law. If you received this e-mail in error, you are requested to destroy the original and any copies of this e-mail in your possession, including any copies which may reside on your hard drive or network server.

[rbrtio]

I wouldn’t call being in the top 1% cheating - it means careful and efficient placement with quality materials. As long as it’s not over 36 eirp, then it’s ok. The real problem is that if you have a high gain antenna, 100 ft AGL, and a good high elevation, we can get hotspots at 50-60 + miles. The issue is that the algorithm in the Helium PoC still doesn’t factor in higher gain antenna or higher elevation/tower placement. So we see invalid witnesses as we are still reaching 50 miles with -115 RSSI. The Helium PoC formula needs to be adjusted to factor in real RF scenarios, which include higher gain antennas and higher AGL placement. Get Outlook for iOShttps://aka.ms/o0ukef … ____ From: curiousfokker @.> Sent: Tuesday, July 13, 2021 3:36:29 PM To: helium/miner @.> Cc: Jeff Hunter @.>; Comment @.> Subject: Re: [helium/miner] Tracking issue: Update RF acceptance curves (#618) Hi Jon, don't worry about it too much. You don't get punished once you get into the top 1%. I think para1 was saying that just because you're in the top 1% doesn't mean you're a cheater. We had serious problems early on with cheaters/gamers. It's almost always worth it to optimize your setup, although keep in mind that location is more important (by far) than elevation or antenna. — You are receiving this because you commented. Reply to this email directly, view it on GitHub<#618 (comment)>, or unsubscribehttps://github.com/notifications/unsubscribe-auth/AUMNYLDK3SXXOAODISQK2GLTXSIT3ANCNFSM4XHZ3RUA. -------------------------------------------------------- This e-mail, and any attachments, are a private communication and contain confidential information intended solely for the use of the person(s) and/or entity to which this e-mail is addressed. If the intended recipient is a client, this communication may also be subject to an internationally enforceable non-disclosure agreement. If you are not the intended recipient, or an agent of the intended recipient, any copying or distribution of this communication is strictly prohibited by law. If you received this e-mail in error, you are requested to destroy the original and any copies of this e-mail in your possession, including any copies which may reside on your hard drive or network server.

The PoC formula is being changed per my comment above. In PoC v11 antenna gain that has been asserted will be taken into account as part of the calculation, however height will not as far as I am aware. In addition the SNR portion is being removed as well. It won't be perfect, but it will certainly help and I can remove the attenuators I have on one of my installations. :-)

[PipArmy]

I bought a miner, 6dbi antenna, and LMR400 cable. I do not understand how 50% of all witnesses and witnessing is invalid. Something is wrong. It’s happening at 3 different locations, 3 different miners.

On Tue, Jul 13, 2021 at 3:36 PM curiousfokker @.***> wrote:

Hi Jon, don't worry about it too much. You don't get punished once you get into the top 1%. I think para1 was saying that just because you're in the top 1% doesn't mean you're a cheater. We had serious problems early on with cheaters/gamers. It's almost always worth it to optimize your setup, although keep in mind that location is more important (by far) than elevation or antenna.

— You are receiving this because you commented. Reply to this email directly, view it on GitHub https://github.com/helium/miner/issues/618#issuecomment-879347876, or unsubscribe https://github.com/notifications/unsubscribe-auth/AGZ6ZUG3HWAJJVUPLO2OB5TTXSIT3ANCNFSM4XHZ3RUA .

[JonPike1]

Ok... Good to see the concern, and thanks for the responses.

The real problem is that if you have a high gain antenna, 100 ft AGL, and a good high elevation, we can get hotspots at 50-60 + miles. The issue is that the algorithm in the Helium PoC still doesn’t factor in higher gain antenna or higher elevation/tower placement. So we see invalid witnesses as we are still reaching 50 miles with -115 RSSI. The Helium PoC formula needs to be adjusted to factor in real RF scenarios, which include higher gain antennas and higher AGL placement.

Oh, yeah... But is it as much the RSSI, vs the SNR at a particular level? It's not that you get a 50mi station at -115. Heck, just checked my latest beacon, and I have a couple at 56km (that long, 140km guy didn't hear me this time) that are -95 and -96, with 0.2 and -7.8 SNR. This on the RAK 1.2dbi stick, 25' up. Odd thing is, the ones that fail are typically 5-10km away, and closer ones don't. The one fail this time, -107dbm, +8 SNR, 6.9km.

I could supply more info if it would be useful.

[mfalkvidd]

however height will not as far as I am aware

It will probably be good to not include height until there is a clear definition whether the height parameter is height or elevation. See https://github.com/helium/helium-wallet-rs/issues/174

The data currently on the chain is probably a mix of height and elevation.

[JonPike1]

Well, does disqualifying witnesses who I hear punish, or not?

On Tue, Jul 13, 2021, 12:36 PM curiousfokker @.***> wrote:

Hi Jon, don't worry about it too much. You don't get punished once you get into the top 1%. I think para1 was saying that just because you're in the top 1% doesn't mean you're a cheater.

[JonPike1]

Agreed. And only AGL makes sense, since it is what really matters to signal coverage.

On Wed, Jul 14, 2021, 1:04 AM Mikael Falkvidd @.***> wrote:

however height will not as far as I am aware

It will probably be good to not include height until there is a clear definition whether the height parameter is height or elevation. See helium/helium-wallet-rs#174 https://github.com/helium/helium-wallet-rs/issues/174

The data currently on the chain is probably a mix of height and elevation.

[kwatkins99]

I think that is debatable. If a hotspot is in a low spot and has a high mast, does it offer the same coverage as someone in a high spot with the same length mast? I would say that probably close to all assertions are AGL. Most people don't know their elevation and then to add the AGL to that. I think we could do without the height variable.

[JonPike1]

Sure... now you've gone and brought the real world into it... ;-)

If we all were on a totally flat plain... then all AGL numbers would be equal, but of course that's not the case, unless you ARE out on the Great Plains or some such. AGL is useful for a first order of approximating range improvement, your terrain may vary... It gets too hard too fast to even think about factoring in terrain variation.

And my comment should have said "Between elevation and AGL, AGL is the only one that is important..."

Wonder what the saner suggestions are for SNR and other stuff.

[PipArmy]

witness_rssi_below_lower_bound Distance 1.3 km RSSI -119 dBm SNR 2.8 dB What am I doing wrong? Again I bought a Bobcat, lmr400 10ft and reduced my height to 8' off the ground and my witness is still invalid. I cannot do anything to fix this. I have now tried 6 antennas. I am using a RAK 5.8dbi

On Wed, Jul 14, 2021 at 6:49 PM Jon Pike @.***> wrote:

Sure... now you've gone and brought the real world into it... ;-)

If we all were on a totally flat plain... then all AGL numbers would be equal, but of course that's not the case, unless you ARE out on the Great Plains or some such. AGL is useful for a first order of approximating range improvement, your terrain may vary... It gets too hard too fast to even think about factoring in terrain variation.

And my comment should have said "Between elevation and AGL, AGL is the only one that is important..."

Wonder what the saner suggestions are for SNR and other stuff.

— You are receiving this because you commented. Reply to this email directly, view it on GitHub https://github.com/helium/miner/issues/618#issuecomment-880259513, or unsubscribe https://github.com/notifications/unsubscribe-auth/AGZ6ZUFPF2SD6AN6YX5EY63TXYIATANCNFSM4XHZ3RUA .

[richardhundt]

Same issue with a bobcat 300 and stock (4dbi) antenna:

Distance 1 km RSSI -119dBm SNR 3.2dB

In the attic of a 2 story home, sparse residential area, on the outskirts of town, interspersed with open fields, so not much noise, and 3 hotspots nearby.

What's interesting about this transaction, is that the beaconer was witnessed by 2 other hotspots 17km away. All 3 hotspots reported an RSSI of -119dBm, despite their distances ranging from 1km - 17km.

One of the 17km witnesses is 77m up, on a tall building, and the other is 42m up. Those guys can see all over the city, so they're not spoofing. Mine's right next door to the beaconer.

What are the chances of having the same RSSI of -119dBm?

My guess is that the RSSI from one of the witnesses clobbers the rest, and that you're not recording them correctly in the transaction.

[PipArmy]

witness_rssi_below_lower_bound Distance 1.3 km RSSI -119 dBm SNR 2.8 dB

What am I doing wrong? Again I bought a Bobcat, lmr400 10ft and reduced my height to 8' off the ground and my witness is still invalid. I cannot do anything to fix this. I have now tried 9 antennas. I am using a RAK 5.8dbi

On Thu, Jul 15, 2021 at 10:54 AM Joseph johnson @.***> wrote:

witness_rssi_below_lower_bound Distance 1.3 km RSSI -119 dBm SNR 2.8 dB What am I doing wrong? Again I bought a Bobcat, lmr400 10ft and reduced my height to 8' off the ground and my witness is still invalid. I cannot do anything to fix this. I have now tried 6 antennas. I am using a RAK 5.8dbi

On Wed, Jul 14, 2021 at 6:49 PM Jon Pike @.***> wrote:

Sure... now you've gone and brought the real world into it... ;-)

If we all were on a totally flat plain... then all AGL numbers would be equal, but of course that's not the case, unless you ARE out on the Great Plains or some such. AGL is useful for a first order of approximating range improvement, your terrain may vary... It gets too hard too fast to even think about factoring in terrain variation.

And my comment should have said "Between elevation and AGL, AGL is the only one that is important..."

Wonder what the saner suggestions are for SNR and other stuff.

— You are receiving this because you commented. Reply to this email directly, view it on GitHub https://github.com/helium/miner/issues/618#issuecomment-880259513, or unsubscribe https://github.com/notifications/unsubscribe-auth/AGZ6ZUFPF2SD6AN6YX5EY63TXYIATANCNFSM4XHZ3RUA .

[lebonez]

Figured I'd chime in I am definitely falling into that between the US915 red and green/orange curve for 5ish witnesses for hotspots within 500m-2km. Losing out on lots of helium since I only receive a few witness further away. Yesterday had a total of 5 invalid. This just started getting worse after moving my hotspot up further in elevation. Hope this gets fixed soon.

[hexalys]

Talked with a member of the Helium team on discord. They are working on a fix to the current RF curve that should help those of us with the invalid witness problem. No word on a release but they are definitely aware.

yup, we're still working on it, just been a little slow due to work on validators and other high priority issues. we're also looking to implement different curves per region - due to differing characteristics like spreading factor, power output, and frequency - which adds a little more complexity.

I have also been getting too many invalid witnesses on a daily basis (up to 5-10 a day), which is frustrating.

One aspect worth pointing out, is inconsistencies with the same hotspot. In my case, a hotspot 511m away, beaconing daily, is valid only 50% of the time, and therefore invalid half the time.

To resolve this: Wouldn't there be a workable way to ensure that once a hotspot has been witnessed by another once (or twice), the constraint can be dropped or greatly minimized. If the distance is the same and no location or antenna reassertion has been made. I don't see why it should be passing a test, proven to have false positives, all the time.

[lebonez]

I think that would open up gaming the system even easier. It just sucks that legitimate hotspots like ours get penalized for having a great setup. Had to downgrade to the 1.2dbi antenna but now I'm missing at least 5 witnesses a day but am getting the closer ones now.

[fletcherMSE]

Just want to say there is hotspot 1.5km from me where we essentially see each other across a wooded valley. Nearly always witness each other, but get invalid witness with some frequency with high SNR:

RSSI -100dBm SNR 9.2dB

Would love it if the acceptance curves could be adjusted to permit/recognize actually valid witnesses.

[GP-Colorado]

My first hotspot has been running for a year now. The first hotspot that it witnessed is around 400 meters away. They were consistently in contact for many months.

A month or two ago when I looked over the witness activity, I noticed that my "neighbor" is now an invalid witness.

In order for it to become "valid", I experimented by changing its "asserted" location to a vacant hex a few blocks away.

That to make a legit relationship "valid", I needed to assert an invalid location might support the need to make some adjustments.