Static Test
In the conducted static test, the hardware specified in Table 1 was deployed at a fixed outdoor location, for an approximate duration of four hours. The objective was to investigate the impact of latency on error occurrences during transitions between navigation messages. Table 2 outlines the RTKLib configuration employed for position determination.
Given that satellite positions are computed via interpolation of the navigation messages over time, concerns arose regarding potential discrepancies resulting from latency-induced delays during ephemeris updates. Real-time results were leveraged to scrutinize positional variations over the test duration. Figure 3 depicts the comparative analysis, revealing that latency fluctuations did not induce significant deviations in any of the three positional components when comparing the solution in real-time with the solution in post-processing. Moreover, Table 3 presents a statistical overview of the disparities observed between real-time processing and post-processing methods. The data illustrates that, while maximum peaks exceeding 10 centimetres may occur, the mean and standard deviation indicate minimal divergence between the results obtained through real-time and post-processing. Despite the expectation of no disparities between the two methods in theory, certain factors can contribute to differences.
-
Simulation error: Despite efforts to emulate the real-time environment accurately, post-processing software may overlook certain aspects, potentially leading to discrepancies in results.
-
Latency and synchronization: Real-time processing involves immediate data processing upon arrival, whereas simulation processes data only after it has been entirely recorded. Consequently, disparities may arise due to latency in real-time data processing.
-
Changing environmental conditions: Unforeseen changes in the real environment, such as fluctuations in the ionosphere or electromagnetic interference, cannot be fully replicated in offline simulations, contributing to differences in outcomes.
-
Accuracy of input data: While RTCM messages aim to provide precise real-time correction data, slight variations in the quality or accuracy of this data may influence position results.
Table 3
Static test statistics on real-time versus post-processing solution disparities
| | dE (meters) | dN (meters) | dU (meters) |
| Maximum (abs) | 0.0832 | 0.1185 | 0.1757 |
| Minimum (abs) | 0.0000 | 0.0000 | 0,0000 |
| Standard deviation | 0.0035 | 0.0053 | 0.0072 |
| Mean | 0.0003 | 0.0004 | 0.0006 |
| Median | 0.0000 | 0.0000 | 0.0000 |
Given the minimal disparities observed, attention can be focused on the issue of latency, the communication methodology between the application and server was pivotal in mitigating latency effects, optimizing network capabilities, and yielding latency results well within acceptable thresholds, thereby safeguarding against errors in static positioning. While acknowledging the multifactorial nature of latency, the findings affirm the viability of the communication approach for static position calculations. Consequently, it is concluded that the methodology is promising for static positioning applications.
Kinematic Test: Pedestrian
To assess the impact of latency in kinematic scenarios, a real-world walking test was conducted. Utilizing the hardware and configuration detailed in Table 1, and the RTKLib setup outlined in Table 2, a 15-minute test was executed under consistent walking conditions. The test route followed an oval trajectory encompassing areas with varied visibility, including wooded regions known to potentially affect positioning solutions.
Comparative analysis was performed between real-time positioning results and post-processed data obtained from the same route. As highlighted in preceding sections, the application not only captures real-time RTKLib outputs but also archives raw sensor data for potential post-processing. Figure 4 presents the trajectory comparison, revealing negligible positional disparities, often below one millimetre. The image depicts the complete trajectory on the left, where the disparity between real-time and post-processing solutions is imperceptible. Conversely, the right portion of the image zooms in on a specific area, highlighting the magnitude of the differences between the two methods.
Figure 5 illustrates the latency metrics derived from the test. The blue line represents the duration between raw data transmission to the server and reception of positioning results, while the red line denotes server-side processing time, encompassing the duration from data receipt to result dispatch. Additionally, Table 4 provides a concise summary of latency statistics extracted from the graph.
Table 4
Static test statistics on real-time versus post-processing solution disparities
| | DTL + DRL + CPL (seconds) | CPL (seconds) |
| Maximum (abs) | 0.8960 | 0.0360 |
| Minimum (abs) | 0.1060 | 0.0030 |
| Standard deviation | 0.0947 | 0.0039 |
| Mean | 0.2556 | 0.0105 |
| Median | 0.2240 | 0.0100 |
Table 5
Pedestrian test statistics on real-time versus post-processing solution disparities
| | dE (meters) | dN (meters) | dU (meters) |
| Maximum (abs) | 0.0036 | 0.0070 | 0.0253 |
| Minimum (abs) | 0.0000 | 0.0000 | 0.0009 |
| Standard deviation | 0.0004 | 0.0009 | 0.0034 |
| Mean | 0.0001 | 0.0001 | 0.0006 |
| Median | 0.0000 | 0.0000 | 0.0000 |
Analysing the results from Table 5, minimal differences are once again evident between real-time and post-processing methods, with a maximum peak difference of only 2 centimetres vertically. These slight disparities may be attributed to numerous factors outlined in the static test. However, they further validate the methodology and affirm that it does not introduce significant errors attributable to latency.
Furthermore, considering the nature of the kinematic test, it is important to acknowledge that due to latency, positions can be displaced depending on the velocity of movement. Assuming an average pedestrian speed of 10 km/h and an average latency of 0.2556 seconds, the positional variation due to this speed would be approximately 0.7 meters. Given the utilization of smartphone-based positioning, it can be inferred that the employed methodology consistently delivers acceptable results regarding latency-induced positional discrepancies.
Kinematic Test: Car
To assess the potential negative impact of latency in practical scenarios, a car-based test was conducted, covering approximately 15 km along a highway at an average speed of 120 km/h. Consistent with previous tests, the same hardware and configuration were used, as detailed in Table 1 and Table 2. The comparative methodology remained consistent, wherein real-time positioning data was juxtaposed with post-processed results.
As depicted in Fig. 6, positional variations between real-time and post-processing were minimal, echoing findings from prior tests. The left side of the image displays the complete trajectory, where the disparities between real-time and post-processing are imperceptible. Conversely, the right side of the image zooms in on a specific area to illustrate the magnitude of the differences more clearly.
Figure 7 illustrates latency results like those observed in previous tests, showing higher Data Transmission, Reception, and Cloud Processing times. Furthermore, Table 6 presents latency statistics closely resembling those from preceding experiments. This test has revealed new latency results, which in this instance are higher compared to the previous test, despite it is using the same network. Several factors could contribute to this discrepancy:
-
Network congestion: Variations in network traffic can influence the duration of data transmission between the mobile device and the server. During peak periods of congestion, packets may encounter delays as they navigate through the network.
-
Network latency: The physical distance between the mobile device and the server can introduce latency as data packets travel across the network infrastructure. Additionally, inefficiencies in routing or congestion along the network path can further contribute to fluctuating latency.
-
Packet loss: Inconsistent latency may also stem from packet loss during data transmission. Lost packets may necessitate retransmission, leading to increased latency and variability in overall transmission time.
-
Server processing time: While the graph in Fig. 7 illustrates constant processing time on the server, it is important to consider potential variations due to factors like resource competition, software optimizations, or background tasks running on the server.
-
WebSocket performance: Despite offering a persistent, low-latency connection between the client (smartphone) and the server, WebSocket performance can still be influenced by factors such as network conditions and server responsiveness. Variability in WebSocket performance may contribute to latency fluctuations.
-
Device performance: The performance of the smartphone itself can impact app latency. Factors such as CPU usage, available memory, and background processes on the device can affect the speed of data transmission and reception.
-
Wireless connectivity issues: Mobile devices may encounter fluctuations in wireless signal strength or connectivity problems, leading to intermittent delays in data transmission to the server.
Table 6
Static test statistics on real-time versus post-processing solution disparities
| | DTL + DRL + CPL (seconds) | CPL (seconds) |
| Maximum (abs) | 0.9530 | 0.0340 |
| Minimum (abs) | 0.3510 | 0.0030 |
| Standard deviation | 0.0965 | 0.0038 |
| Mean | 0.5313 | 0.0105 |
| Median | 0.5210 | 0.0100 |
Table 7
Car test statistics on real-time versus post-processing solution disparities
| | dE (meters) | dN (meters) | dU (meters) |
| Maximum (abs) | 0.0071 | 0.0043 | 0.0136 |
| Minimum (abs) | 0.0000 | 0.0008 | 0.1635 |
| Standard deviation | 0.0005 | 0.0004 | 0.0011 |
| Mean | 0.0001 | 0.0000 | 0.0002 |
| Median | 0.0000 | 0.0000 | 0.0000 |
Considering an average latency of 0.5313 seconds at a speed of 120 km/h, the resulting positional variation attributed to latency averages around 17 metres. At these speeds, the positional variation becomes significantly greater. Therefore, in scenarios where latency is higher, it can result in a substantial displacement of the vehicle's position by the time the last calculated position is received. This implies that, depending on the application, such discrepancies could be critical.
Once more, Table 7 presents statistics comparing the three position components between real-time and post-processing methods. As observed in previous cases, although occasional peaks of a few centimetres may occur, the disparity in position between real-time and post-processing remains minor. This reaffirms the validity of the methodology even under high-speed conditions.
As a result, this test highlights that at high speeds, even slight increases in latency values can lead to significant positional variations. This suggests that the methodology may not be suitable for applications where security is paramount, despite yielding results. However, it also underscores the potential for further research aimed at reducing latency within the same methodology, thereby opening avenues for improvement and innovation.
Figure 8 offers a comprehensive visual overview, featuring a sequence of screenshots extracted from the application, depicting the test route conducted on a highway by car. These displayed screenshots furnish detailed insights into the performance of the positioning solution. While the primary intent of these findings is not to assess the accuracy of the solution, given its susceptibility to inherent displacement due to speed, they serve to highlight the continuous nature of the solution. The trajectory exhibits smooth transitions without abrupt deviations.
Moreover, the minimal presence of significant outliers attests to the robustness of the employed positioning algorithm. Notably, the solution's accuracy is commendable, as manifested by the consistent alignment of the route with the actual path traversed, notwithstanding the displacement induced by speed and latency. In summary, the visual representation delineated in Fig. 8 underscores the efficacy and dependability of the application in producing precise positioning solutions under dynamic scenarios, such as vehicular travel on roads. However, it is worth noting, as previously mentioned, that movement at elevated speeds can induce positional shifts attributable to latency.