YERKON A terrestrial positioning backup layer Run the simulation

The comparison table

Thirteen systems side by side under the same columns. The three shaded rows are the YERKON rows the simulation filled; the rest are what their own sources publish. What was done to a cell is in the box the button at the end of its row opens and in the notes at the foot of the page, and an empty cell means the source publishes nothing that fits that column.

What the accuracy costsWhat the accuracy costs02,5 k5,0 k7,5 k10 k02,557,51012TL/km²HPE P95 [m]Galileo OSBeiDou BDSGPSYERKON (Kırsal)YERKON (Şehir içi)
What the accuracy costsWhat the accuracy costs05,0 k10 k02,557,51012TL/km²HPE P95 [m]GalileoBeiDouGPSKırsalŞehir içi
  • YERKON: what this simulation found
  • Other systems: the figure their own sources publish
  • A ceiling: the true value lies below the mark
Capital per square kilometre across, horizontal error up, both from nought. Left and down is better: cheap and precise. The satellite systems are divided by all the land on earth, YERKON by each row's own area. YERKON's rural row costs less than the satellite systems and its town row more: in town buildings cut the signal, so a square kilometre needs more units. Both sit in the satellites' band of accuracy. The tunnel row is absent: it is divided by route kilometre and does not belong on this axis. A system that publishes only one of the two cannot be drawn either.

Accuracy side by side

Horizontal error, worst 5 % excluded (HPE P95)Horizontal error, worst 5 % excluded (HPE P95)0,1110meLoran15,72YERKON (Şehir içi)9,19YERKON (Kırsal)8,15GPS≤ 8 / ≤ 15TerraPoiNT≤ 8GLONASS≤ 5 / ≤ 12YERKON (Tünel)2,43QZSS SLAS≤ 1,70BeiDou BDS≈ 1,52Galileo OS≤ 1,41Pozyx0,181Locata0,017
Horizontal error, worst 5 % excluded (HPE P95)Horizontal error, worst 5 % excluded (HPE P95)0,1110meLoran15,72Şehir içi9,19Kırsal8,15GPS≤ 8 / ≤ 15TerraPoiNT≤ 8GLONASS≤ 5 / ≤ 12Tünel2,43QZSS≤ 1,70BeiDou≈ 1,52Galileo≤ 1,41Pozyx0,181Locata0,017
  • YERKON: what this simulation found
  • Other systems: the figure their own sources publish
  • A ceiling: the source says "at most this", the true value lies the way the arrow points
  • Two figures: the dot on the average, the end of the line on the worst place
Further left is better. The axis is logarithmic because the figures run from centimetres to fifteen metres. A system with an empty cell is not drawn.
Capital per square kilometreCapital per square kilometre02,0 k4,0 k6,0 k8,0 k10 kTL/km²YERKON (Şehir içi)9061BeiDou BDS≥ 2714,05GPS≈ 2341,76Galileo OS≈ 2109,23YERKON (Kırsal)1138
Capital per square kilometreCapital per square kilometre02,5 k5,0 k7,5 k10 kTL/km²Şehir içi9061BeiDou≥ 2714,05GPS≈ 2341,76Galileo≈ 2109,23Kırsal1138
  • YERKON: what this simulation found
  • Other systems: the figure their own sources publish
  • A floor: the source says "at least this", the true value lies the way the arrow points
The systems that publish a capital cost. The satellite rows are divided by all the land on earth, the YERKON rows by their own area. The tunnel row is absent: it is divided by route kilometre rather than by square kilometre, so it does not belong on this axis. Its note in the table says so.

What the columns mean

ColumnMeaning
HPE, VPEHow far the position is out, horizontally and vertically, in metres. P50 is the value half the errors stay under, P95 the one 95 per cent of them stay under.
AvailabilityWhat share of the attempts to work out a position gave a valid one. It counts only the failures in the model, and is not a promise about a service.
AreaThe ground where enough units can be heard to work out a position, found by sweeping the real terrain. Not the same as the ground the signal reaches.
CAPEXUnit hardware cost divided by the service area; every part of the unit and the fitting included.
OPEXWhat a year of running one km² costs, from an itemised list of what recurs.

Three warnings for reading the table

SystemTechnologyEnvironmentHPE P50 [m]HPE P95 [m]VPE P95 [m]Availability1Area [km²]CAPEX [TL/km²]OPEX [TL/km²/yr]This row's sources and notes
GPSGNSSDış-2≤ 8 / ≤ 153≤ 13 / ≤ 33≥ %99 / ≥ %90≈ 148,94 milyon4≈ 2341,765≈ 130,106
Galileo OSGNSSDış-≤ 1,417≤ 2,40≥ %99,588≈ 148,94 milyon≈ 2109,239≈ 223,0110
GLONASSGNSSDış-≤ 5 / ≤ 1211≤ 9 / ≤ 2511≥ %99 / ≥ %9011≈ 148,94 milyon-12≈ 61,1013
BeiDou BDSGNSSDış-≈ 1,5214≈ 2,6414> %9515≈ 148,94 milyon≥ 2714,0516-17
QZSS SLASBölgesel GNSS düzeltme servisiDış-≤ 1,7018≤ 2,7318%99,99619≈ 377974,8720-21≤ 10904,4922
NavIC SPSBölgesel GNSSDış--23-24--25-26-
TerraPoiNTKarasal konumlandırma, UHFİç + dış≤ 327≤ 827228%99,9628≈ 90029-30-
LocataKarasal pseudolite konumlandırmaİç + dış-0,017310,02431%10032≈ 647533-34-
PozyxUWB RTLSİç + dış0,114350,18135--36≥ 0,03537-38-
eLoranKarasal düşük frekanslı konumlandırmaDış-15,7239-40-41-42-43-
YERKON (Şehir içi)44Karasal konumlandırma (SX1280/LoRa TWR)Dış2,299,194,00%96,65458,14906146194046
YERKON (Kırsal)47Karasal konumlandırma (E28-SX1280 TWR)Dış2,168,154,96%94,5845354,0011384633046
YERKON (Tünel)48Karasal konumlandırma (UWB/DWM3000 TWR)İç + dış0,582,433,31%94,32450,02108294 /km464922107 /km4649

Run on 2026-09-29 against defaults.toml. The shadows were drawn 8 times and the ground profile was read every 10 m. The first three columns are the names that go into the Turkish report, so they are not translated.

Notes (49)

  1. Availability is the share of time or of situations in which a valid position solution meeting the stated accuracy and service conditions can be provided. The sources use different test durations, regions and performance thresholds, so these percentages must not be compared as though they came from one protocol. ↑
  2. No current, comparable P50 user position error measured the same way is published for GPS, Galileo, GLONASS or BeiDou. No P50 was derived from a P95. GPS.gov, GPS performance · Galileo Open Service performance report, Q2 2026 ↑
  3. The GPS figures were checked again against the 2024 GPS SPS performance analysis. The first figure is the global average and the second the worst location: 95 % horizontal error under 8 m and 15 m, vertical under 13 m and 33 m, availability over 99 % at the average location and over 90 % at the worst. The 2024 analysis found the L1 C/A commitments met. GPS.gov, GPS performance · GPS.gov, performance standards ↑
  4. For the global satellite systems the area denominator is the earth's land surface with the oceans and seas taken out, about 148,94 million km² (148940000 km²). The earth's whole surface, oceans included, is about 510,06 million km²; land is about 29 % of it. GPS, Galileo, GLONASS and BeiDou serve the sea as well; the choice does not limit their service to land, it only puts them on the same denominator as YERKON and the other terrestrial systems, which are built on land. Nor can a single area where GNSS certainly fails on land be subtracted: tunnels, underground structures, the inside of buildings, deep street canyons, terrain shadow, dense vegetation, multipath and local interference all degrade the solution, but how much depends on the receiver, the antenna, the satellite geometry and the time. CIA World Factbook, the world's land and water area (archived copy, 28 April 2025; the Factbook was retired on 4 February 2026) · NASA, the Moon by the numbers (Earth column: surface area with the oceans, 510064472 km²) ↑
  5. The GPS CAPEX figure is a historical investment proxy derived from 7,2 billion USD of procurement plus research and development. It is neither today's cost of building it again nor an accounting CAPEX. It was converted to lira at the 4 September 2026 rate and divided by the land denominator. National Academies, historical GPS investment · Banque de France, reference rates of 4 September 2026 ↑
  6. The GPS OPEX figure is the 1993 DOD/DOT joint task force's estimate of sustaining a minimum GPS service, about 400 million USD a year in FY1993 dollars. No inflation adjustment was applied. U.S. Government GPS Archive, yearly sustainment cost · Banque de France, reference rates of 4 September 2026 ↑
  7. The Galileo cells hold the complementary measurement in the Q2 2026 Open Service report: Galileo only, dual frequency, 3D, including propagation and user equipment error, 95 % horizontal error 1,41 m and vertical 2,40 m over the quarter. That measurement is not a minimum performance commitment; 7,5 m and 15 m are not measured errors but the error conditions of the positioning availability assessment. The limits committed for the system's own contribution are 5 m horizontal and 8 m vertical on dual frequency and 10 m and 16 m on single frequency; the achieved values are well inside them. Galileo Open Service performance report, Q2 2026 · European GNSS Service Centre, Galileo OS performance ↑
  8. In Q2 2026 Galileo's positioning availability was 99,58 % at the worst user location in April and 99,98 % at the average one; both were 100 % in May and June. The cell holds the lowest result of the quarter. The official minimum targets are 87 % at the worst location and 90 % at the average one, assessed against 7,5 m horizontal and 15 m vertical. Galileo Open Service performance report, Q2 2026 ↑
  9. The Galileo CAPEX figure is an investment proxy derived from the 5,58 billion EUR historical programme implementation cost, which covers development and deployment. European Court of Auditors, Galileo deployment cost · Banque de France, reference rates of 4 September 2026 ↑
  10. The European Commission's 2011 cost table gives the average yearly cost of operating the full infrastructure, in 2010 prices, as 800 million EUR: 590 million for Galileo, 110 million for EGNOS and 100 million for other items. Only Galileo's 590 million was used. It is a historical estimate in 2010 prices, not what was spent in 2026. European Commission, Galileo operating cost · Banque de France, reference rates of 4 September 2026 ↑
  11. For GLONASS the first figure is the average and the second the worst location performance. No newer global measurement published with the same definition and statistic was found, so the standard's figures were kept. ICAO Annex 10, GLONASS and BeiDou availability · GOST 32454-2013, GLONASS performance thresholds ↑
  12. The GLONASS CAPEX cell is left empty. The public historical programme figures do not reliably separate development, use and sustainment from pure deployment investment. Kommersant, GLONASS deployment period cost ↑
  13. The GLONASS OPEX figure is a historical proxy: the simple nine year average of the 146898,9 million RUB system sustainment budget for 2012 to 2020. Kommersant, GLONASS operating programme costs · Kommersant, GLONASS programme cost items · Banque de France, reference rates of 4 September 2026 ↑
  14. The BeiDou cells hold official achieved performance: by the China Satellite Navigation Office's iGMAS assessment, B1C single frequency global 95 % horizontal accuracy is about 1,52 m and vertical about 2,64 m. The service standard's limits are 10 m horizontal and vertical globally, 5 m in Asia Pacific. BeiDou, B1C global 95 % performance (iGMAS) · BeiDou Open Service Performance Standard 2.0 ↑
  15. The BeiDou Open Service Performance Standard 2.0 defines system service availability as over 95 %. The figure the same source puts over 0,99 is satellite availability; it is not the same measure as positioning service availability, so it was not put in the cell. BeiDou Open Service Performance Standard 2.0 · BeiDou, B1C global 95 % performance (iGMAS) ↑
  16. The BeiDou CAPEX figure is a proxy that takes the lowest historical programme estimate CASI reports, 56 billion CNY, as a lower bound. It is not an accounting CAPEX. CASI, BeiDou programme cost estimate · Banque de France, reference rates of 4 September 2026 ↑
  17. No current yearly operating total for the BDS-3 global constellation and its main ground segment was found. Regional CORS operating tenders were not used as a stand-in. CASI, BeiDou programme cost estimate ↑
  18. The QZSS SLAS figures are not service design limits but the monthly 95 % results in the FY2025 second half performance report: over 13 evaluation points from October 2025 to March 2026, the highest published horizontal P95 is 1,70 m (Itoman) and the highest vertical P95 2,73 m (Hiroshima). They are not one 3D error pair that occurred together. QZSS, SLAS performance report for the second half of FY2025 ↑
  19. The constellation service availability measured in the FY2025 second half is 0,99996. The official minimum specification is 0,9997. QZSS, SLAS performance report for the second half of FY2025 ↑
  20. The SLAS service area is 13 circles of 100 to 350 km radius, but their combined area is not published in km² and much of it is sea. For a land comparison Japan's national area of 1 April 2026, 377974,87 km², is used as an open proxy. It is not the real SLAS service area. QZSS, SLAS performance report for the second half of FY2025 · Geospatial Information Authority of Japan, national area on 1 April 2026 ↑
  21. The share of satellites, ground system, launch, development and long service contracts that belongs to SLAS alone cannot be separated out of public data. Cabinet Office Japan, QZSS programme cost ↑
  22. 13,294 billion JPY a year is the whole FY2026 budget for providing the QZSS service, not SLAS alone. It was divided by the Japanese land proxy, so it is an upper bound and is written with "≤". Cabinet Office Japan, QZSS FY2026 budget · Cabinet Office Japan, QZSS budget archive · Banque de France, reference rates of 4 September 2026 ↑
  23. The NavIC row is empty to reflect its current state. By the Indian government's statement of 29 July 2026 only three satellites providing position are working: IRNSS-1B, IRNSS-1I and NVS-01. An independent position solution needs at least four; the timing service continues. Government of India, current NavIC status ↑
  24. The under 20 m historically given for NavIC is 2σ user position accuracy, not a P95, so it was not put in the P95 columns. ISRO, NavIC accuracy and approximate service area · ISRO, NavIC service area limits ↑
  25. No land area is published for NavIC. The rectangle from ISRO's approximate latitude and longitude limits (33243459 km²) takes in large areas of sea, so it is not used for a land comparison. ISRO, NavIC service area limits ↑
  26. The historical 1420 crore INR budget covers seven orbital satellites, two ground spares and the ground segment, and leaves launch out. With no current independent service and no suitable land denominator it was not turned into TL/km². Government of India, NavIC deployment cost · Government of India, what the NavIC cost covers ↑
  27. TerraPoiNT's horizontal P50 and P95 come from one 2D navigation test published at ION GNSS+ 2021. Figures from other tests, around 9 to 14 m, are not in this cell. ION, NextNav TerraPoiNT P50/P95 2D performance ↑
  28. The 2 m vertical and 99,96 % availability come from the JRC's alternative positioning assessment, not from the ION horizontal test. The availability is under the JRC's assumptions and must not be read as a commercial service commitment. European Commission JRC, NextNav vertical accuracy ↑
  29. About 900 km² is the reported scale of the network of about 100 sites in the San Francisco Bay Area, not the technology's maximum reach. NextNav later built networks elsewhere, but no single total km² with the same accuracy and cost scope has been published. SEC, NextNav investor day, Bay Area network scale · SEC, NextNav deployments ↑
  30. Neither the full build cost nor the full yearly operating cost of that same 900 km² network is published. SEC, NextNav investor day, Bay Area network scale ↑
  31. Locata's 0,017 m horizontal and 0,024 m vertical are the position result in the JRC performance assessment. European Commission JRC, Locata performance ↑
  32. The availability measured in the JRC's one day demonstration is 100 %. The 99,9999 % figures are projections for 14 and 100 day GNSS outages, not measured long term uptime. European Commission JRC, Locata performance ↑
  33. About 6475 km² is the reported White Sands installation of about 2500 square miles in km². It is not the installation the JRC measured; it is only an example of real deployment scale. Inside GNSS, the Locata White Sands contract · Airforce Technology, Locata White Sands coverage ↑
  34. The White Sands installation is made of several contracts and integration items; no full deployment cost can be drawn from one figure. Inside GNSS, the Locata White Sands contract ↑
  35. Pozyx P50 0,114 m and P95 0,181 m are the median and 95th percentile measured on a figure of eight path in a peer reviewed study. They are not a general product guarantee. Sensors, Pozyx P50/P95 position error ↑
  36. The over 99,95 % uptime in the case study is a technical requirement rather than measured availability, so the cell is empty. Pozyx, a 35000 m² installation with 250 anchors ↑
  37. 0,035 km² comes from a real industrial installation in 2023, the larger of two European sites, over 35000 m² with 250 anchors. Pozyx, a 35000 m² installation with 250 anchors ↑
  38. Catalogue hardware and licence prices do not represent the full contract and installation cost of the 35000 m², 250 anchor project. Pozyx, a 35000 m² installation with 250 anchors ↑
  39. eLoran's 15,72 m is the 95 % horizontal position accuracy reported within 30 km of Incheon Port with ASF correction. MDPI Remote Sensing, eLoran 95 % horizontal accuracy and ASF ↑
  40. Standalone eLoran gives no altitude in a 2D solution, so VPE P95 does not apply. MDPI Remote Sensing, eLoran 95 % horizontal accuracy and ASF ↑
  41. The 99,7 % for eLoran is a requirement threshold rather than availability measured on the Korean testbed, so the cell is empty. Korean journal of positioning, navigation and timing, eLoran testbed performance requirements ↑
  42. Assuming three circles of 30 km radius do not overlap at all gives a theoretical upper bound of 8482,30 km²; it is not used. One circle's geometric area is about 2827,43 km², but the source does not show 15,72 m being delivered at every point in it, so that is not in the cell either. Korean journal of positioning, navigation and timing, the current Korean eLoran installation · South Korea's national maritime office, the eLoran architecture ↑
  43. The project budgets and individual maintenance tenders are not a full deployment or national yearly operating total for the area where the accuracy was measured. Republic of Korea MOF, eLoran project budget ↑
  44. The urban row was run over the real ground at Kızılay; the units drive the real streets. Standing lighting columns, traffic lights, signs and stops, and masts to be put up on hilltops, were all tried as places for the units; the search chose 19 lighting columns among them. The placement in the simulation is there to show cost feasibility: with more units, or units elsewhere, the accuracy wanted can be had where it is wanted. Its range error model is derived from published SX1280 measurements, so the result is a model output rather than a field measurement. SX1280 datasheet · SX1280 long range test · YERKON on GitHub, the simulation code and its method ↑
  45. YERKON availability is the share of ranging rounds whose position came out reliable enough to hold the 10 m target. The filter that works out the position also estimates its own error; where that estimate is over 5,78 m across, the round gives no position. 5,78 m is the bound that matches 95 % of errors within 10 m: 10 / 2,45 = 4,08 m on each axis, 4,08 x 1,41 = 5,78 m for both together. The service availability the satellite providers publish rests on different test definitions, and the percentages must not be read as the same metric. YERKON on GitHub, the simulation code and its method ↑
  46. YERKON CAPEX is the thousand unit bill of materials price of a scenario's broadcast units plus what fitting each one costs (bracket, labour and, where needed, a standalone supply), divided by the scenario's area. The unit price covers every part of the board, its box, the printed board and its assembly; test, calibration, certification, tax and shipping are excluded. The position comes from a Kalman filter that combines radio ranges with the height from a map; no inertial unit or odometry is used. YERKON OPEX is the yearly running cost: electricity (the commercial tariff, 5,62 TL/kWh after the April 2026 increase), replacement (tax depreciation lives), maintenance and the central system. A maintenance visit is a crew day (bucket truck, electrician and helper, 17570 TL) divided by the sites the crew reaches on that structure in a day: 8 lighting columns in town, 4 distribution poles in open country, 4 units in a tunnel. The central system is one day a week of a developer and two servers, 237000 TL a year spread over 1000 units. DigiKey, electronic component prices · LCSC, electronic component prices · Mouser, electronic component prices · Banque de France, reference rates of 4 September 2026 · Ankara Sepetli Vinç Kiralama: price guide (a 20 m bucket truck about 2000 TL an hour, 30 % off by the day) · Yapı Arenası: 2026 trade wages (12 August 2026) · Dicle Elektrik, Suriçi: 414 luminaires on 207 poles in one weekend with 11 crews and 11 bucket trucks, 25 June 2026 · forelektrik.com: what a kWh costs, 2026 (commercial, single rate, taxes included 4,78 TL/kWh), 26 December 2025 · Milliyet Uzmanpara: the April 2026 electricity increase (low-voltage public and private services 17,5 %), 4 April 2026 · Yazılımcı Maaşları 2026 survey (1223 respondents, net median 95000 TL at companies in Türkiye), February 2026 · Karekod: cheapest VDS prices 2026 (6 vCPU, 8 GB, 374,90 TL a month), 25 September 2026 ↑
  47. In the rural row the units drive Polatlı's real roads; the 52 anchors stand where a placement search put them (47 distribution poles, 5 lighting columns). Distribution poles along the roads, standing lighting columns and signs, and masts to be put up on hilltops were tried; roofs are rented, so the search is not offered them. The placement in the simulation is there to show cost feasibility: with more units, or units elsewhere, the accuracy wanted can be had where it is wanted. Its error model is derived from published SX1280 measurements at short range. Its long range result is therefore a model output rather than a direct field measurement. The distribution pole rent is taken as nothing: the units go up under a public-sector agreement, with the prospect of being included in public structures that already stand, such as ITS cabinets. Where the road has an ITS point the unit is fitted to it; elsewhere it runs on its solar panel and passes its messages for the centre by its own radio to the nearest unit with a line. SX1280 long range test · YERKON on GitHub, the simulation code and its method ↑
  48. The tunnel row is worked out with the DWM3000 on channel 5 (6489,6 MHz): the BTK location tracking limit (-41,3 dBm/MHz), Qorvo's channel 5 sensitivity (-100 dBm) and the module's own antenna (0 dBi, as no gain is published). The loss in the bore follows the model Molina-Garcia-Pardo and colleagues measured in a road tunnel at 2,8 to 5 GHz between 50 and 500 m (equation 2), carried to 6,5 GHz rather than measured there. The vehicle's UWB device may send less power above its own height (-53,3 dBm/MHz, ETSI EN 302 065-3, 4.3.4.2); units below the vehicle's antenna, 1,2 m above the road, stay clear of that limit. The result is a Monte Carlo simulation rather than a field measurement. Qorvo DWM3000 datasheet · Qorvo DW3000 Data Sheet Rev 1.3, receiver sensitivity (tables 7, 8 and 9) · Molina-Garcia-Pardo, Lienard, Degauque: Propagation in Tunnels, EURASIP JWCN 2009 · BTK: technical criteria for radio equipment exempt from frequency assignment (Board decision 23.09.2022, 2022/İK-SYD/245) · ETSI EN 302 065-3 V2.1.1: UWB in road and rail vehicles · YERKON on GitHub, the simulation code and its method ↑
  49. The tunnel row's two cost cells are divided by route kilometre rather than by square kilometre, and are marked "/km". A 2 km tunnel 12 m wide covers a fiftieth of a square kilometre; dividing by something that small makes the number large because the denominator is small, not because a tunnel is dear. A tunnel serves a line rather than an area, and cost per kilometre is the figure a tunnel operator would ask for. The other two rows are divided by area, so this cell is not the same measure as theirs. ↑