GNSS dilution of precision (DOP) describes how a selected satellite geometry transforms assumed measurement uncertainty into position or clock uncertainty. It does not measure the error of the current position. To interpret a DOP value, identify the observations and model behind it, then examine their errors and the resulting solution separately. More satellites can improve geometry without making every added measurement trustworthy.
For a GNSS integration engineer, this distinction answers three different questions: whether the directions are useful, whether the measurements deserve their assigned weight, and whether the resulting coordinates meet the task. A single green DOP indicator cannot answer all three.
Geometry changes how range errors combine
Each pseudorange constrains the receiver along a satellite-to-receiver direction. Measurements from similar directions provide similar constraints; more diverse directions help distinguish changes in the different position components. Receiver clock offset is another unknown shared by the observations, so a useful geometric arrangement must also separate position from clock effects.
Navipedia's code-based positioning model linearizes the range equations around an approximate position. Its geometry matrix, G, relates small changes in coordinates and receiver clock offset to changes in the predicted ranges. This is more informative than counting satellites: the directions determine the matrix, not just the number of its rows.
For an unconstrained three-dimensional solution with one receiver-clock unknown, G needs four independent columns. Four observations are therefore a minimum count, not a guarantee of a well-conditioned solution. Additional clock states, constraints or other measurements change the model; the simple four-unknown description is not every receiver's implementation.
HDOP and PDOP describe different components
HDOP concerns horizontal position geometry. VDOP concerns vertical geometry. PDOP combines the three position components. For the same observation set and conventional definition, PDOP² = HDOP² + VDOP². Trimble's PDOP reference gives this relationship and discusses geometry-based collection planning.
These values are dimensionless. HDOP 1.2 is not a horizontal error of 1.2 metres, and comparing one receiver's HDOP with another's PDOP does not rank their accuracy. GDOP additionally includes the clock component under its stated convention; it should not silently replace a position-only measure.
A planning DOP is useful because geometry can be calculated before measurements are collected. That advantage also explains its limit: the calculation cannot discover a local reflected path or an unexpected measurement bias merely from where the satellites are predicted to be.
A conversion to metres needs an error model
Navipedia's positioning-error treatment separates formal, predicted and measured accuracy. In its simple zero-mean, independent, equal-variance range-error model:
Q = (GᵀG)⁻¹; predicted 3D position RMS = σρ × PDOP.
Here σρ is the assumed standard deviation of each range error, in metres; PDOP comes from the position part of Q. The matrix must be full rank. The result is a model prediction, not the measured error at an epoch or a guaranteed radius around the receiver.
Consider an invented comparison using that model, not receiver test data. PDOP 2.0 with σρ = 0.5 m predicts 1.0 m three-dimensional RMS. PDOP 1.5 with σρ = 2.0 m predicts 3.0 m. The smaller DOP has not produced the smaller prediction because the measurement-error assumption changed. These are neither 95% bounds nor ResiNav accuracy figures.
Real observations can have unequal uncertainties, correlations and biases. A weighted linear covariance uses P = (GᵀR⁻¹G)⁻¹, with R representing the assumed measurement-error covariance. A geometry-only DOP and a receiver's weighted uncertainty estimate need not tell the same story. Neither becomes a measurement of truth just because it is reported to several decimal places.
Use the satellites in the solution, not every tracked signal
A receiver can track a satellite without using it in a particular solution. Health, corrections, measurement screening and operating mode can change the used set. Several frequencies from one satellite also do not supply several different satellite directions.
The NovAtel OEM7 RTKDOP documentation explicitly associates that log with satellites used in the RTK solution. It explains why tracked satellites may be unused and notes that its DOP values do not account for the different weights assigned to satellites or signals. Those are NovAtel interface definitions, not assumed ResiNav outputs.
When comparing a DOP trace with a position trace, match the epoch, solution type, used observations and applicable masks. A planning calculation using an open sky is not a substitute for the subset actually retained behind an obstruction. The multi-constellation versus multi-frequency guide explains the separate capability choice; access to more signals is not proof that all of them entered this solution.
More geometry can coexist with more position bias
The following is a constructed linearized teaching calculation, not a live constellation, receiver algorithm or product test. Solve for east, north, up and one clock offset expressed in metres, using equal weights and no additional constraints. Azimuth is measured clockwise from north; elevation is above the horizon. Each geometry row is:
g = [−cos(elevation) sin(azimuth), −cos(elevation) cos(azimuth), −sin(elevation), 1].
Use six assumed azimuth/elevation pairs in degrees: (0,30), (90,30), (180,30), (270,30), (45,60), (225,60). A seventh direction is (135,45). Compute Q from each geometry; independently solve the stated range-error vector for its position and clock effect.
| Assumed observations | Geometry-only PDOP | Result of the stated error vector |
|---|---|---|
| Six directions; every range error set to zero | 2.6009 | Zero position and clock error in this constructed calculation |
| Seven directions; every range error set to zero | 2.5442 | Zero position and clock error in this constructed calculation |
| The same seven directions; only the seventh range has an assumed +10 m bias | 2.5442 | Approximately 4.350 m 3D position error and −0.842 m clock-offset error |
The seventh direction improves the geometry number. Its +10 m bias does not change that number at all, because DOP was calculated from directions rather than the measured ranges. Yet the equal-weight solution responds to the inconsistent observation. The clock result is a range-equivalent offset, not seconds; the position result is not simply the original 10 m bias.
This is also why removing an observation can increase DOP while reducing actual error. In this deliberately known-error example, removing the seventh observation returns to the six-direction result. A real receiver does not automatically know which observation is wrong. Navipedia's RAIM explanation describes the separate use of measurement redundancy and consistency checks; a low DOP is not such a check, and no fault-detection function is assigned to a ResiNav model here.
To interpret the fit itself, distinguish GNSS measurement residuals from actual position error, including errors absorbed into position and clock estimates.
Interpret exclusion without choosing a satellite-count target
Under the same full-rank, unbiased, independent measurement model, adding a correctly modelled observation contributes information; it does not worsen the formal covariance. A biased or misweighted observation violates the premise behind using that result as a prediction of actual error. Conversely, excluding observations reduces geometric information and may leave a weak or unavailable solution.
The useful comparison therefore retains both effects. For matched epochs, examine the used set, documented weighting or exclusion state, model-based uncertainty and errors against an independent reference. Retain intervals with no acceptable solution. Do not reward a method merely for keeping more satellites, or merely for rejecting more of them.
For the settings behind that comparison, see how GNSS elevation masks, C/N0 screening and observation weighting differ, including where each acts and how to compare its consequences.
The multipath and NLOS guide explains one source of biased ranges. This article adds the missing propagation step: how those measurement errors and the selected geometry enter a position estimate, including why a better geometry statistic can accompany a worse result.
Give each reported quantity its own job
Use DOP to interpret geometry under a stated observation model. Use a documented uncertainty output to understand what the receiver predicts under its model. Use a suitable independent reference to measure position error. Use the applicable integrity method and status when the task requires a bound and warning function; the integrity guide covers that separate responsibility.
If the investigation independently establishes a GNSS interference-protection requirement, the ResiNav catalogue is the product-screening route. DOP alone does not establish that requirement or demonstrate a protection benefit. Its value is narrower and useful: it explains the geometric part of the estimate without pretending to measure everything that can make the coordinates wrong.
NEXT STEP
Turn available platform information into an engineering review.
Use the Technology centre to frame the discussion, review relevant application scenarios, then send the available platform and receiver details for confirmation.