Performance Evaluation of a Deep Learning-Enhanced Software-Defined Receiver for IRNSS SPS Signals in Vietnam

During the acquisition stage, the uncertainty regarding the availability of specific satellites and the extent of Doppler shift and code delay require a comprehensive search across all possible values of code and Doppler frequency. This is achieved by generating a local code and carrier signal and then performing a correlation with the received signal. If a satellite is present, a prominent peak will emerge above the ambient noise, as illustrated in PRNs 1, 2, 3, 5, 7, 9, and 10 (Figure 8). In contrast, if no satellite is available, the correlation result will only show ambient noise, as depicted with PRNs 4, 6, 8, 11, 12, 13, and 14.

Figure 8.

Acquisition CAF plots for all IRNSS PRNs at 12:31:37 May 23rd 2024

Various criteria exist to determine whether a satellite has been acquired; for this analysis, instead of relying solely on traditional correlation-based detection, we applied a CNN-based classifier to identify acquisition peaks from correlation maps. This deep learning model was trained offline on synthetic CAF data to handle different noise and Doppler conditions, significantly reducing false positives in noisy channels.

Since the tracking results of the acquired satellites show similar behavior, PRN 10 is selected for visualization. The tracking result of PRN 10 in Figure 9 indicates that the receiver is operating in a low signal-to-noise ratio (SNR) environment, which is reflected in the noisy outputs of both the PLL and DLL discriminators. Low SNR can cause increased fluctuations in the phase and code tracking loops, leading to potential cycle slips in the PLL and a degraded ability to maintain precise code alignment in the DLL. This issue is especially apparent in the raw discriminator outputs, where the higher amplitude noise affects the tracking accuracy.

Figure 9.

Tracking plot for PRN 10 from 12:31:37 to 12:32:27 (50 seconds) on May 23rd 2024

Despite the challenges posed by the low SNR environment, the tracking results demonstrate successful acquisition and tracking of the IRNSS signal. The stable patterns in both the PLL and DLL discriminator outputs, particularly after filtering, indicate that the receiver maintains a reliable lock on the carrier phase and code phase. Although noise is present, extending the tracking loop bandwidth and lengthening the integration time have helped to mitigate its impact, allowing the receiver to maintain consistent tracking.

As shown in Figure 10, the figure illustrates the Carrier-to-Noise Density Ratio (C/N0) for the NavIC L5 band satellites, capturing signal strength variations under low-elevation and low-SNR conditions as observed in Vietnam. The results show that while some satellites maintain relatively high C/N0 values, others display significantly lower levels, leading to varying tracking performance across satellites.

Figure 10.

C/N0 values of all PRNs from 12:31:37 to 12:32:27 (50 seconds) on May 23rd 2024

For satellites with higher C/N0 values, the tracking loops are likely locked, indicating successful acquisition and stable tracking. These satellites, positioned at slightly higher elevations or with less obstruction, provide a sufficiently strong signal for the receiver to maintain a consistent lock.

However, satellites with low C/N0 values, often due to lower elevation angles and increased atmospheric attenuation, exhibit weaker signal strengths. These satellites face frequent unlocks and unstable tracking, as the signal fades below the receiver’s detection threshold.

In this environment, only a subset of satellites is consistently visible and tracked successfully. The remainder, hindered by poor signal quality, cannot maintain the lock reliably, leading to intermittent tracking or complete loss of lock. This variation emphasizes the impact of low elevation and low SNR on satellite visibility and tracking stability, highlighting challenges in using NavIC signals for navigation in fringe coverage areas like Vietnam.

In Figure 11, the sky plot demonstrates that most visible IRNSS satellites are at low elevation angles, typical for Vietnam, which lies at the edge of the IRNSS coverage area. Despite this low elevation, the Position Dilution of Precision (PDOP) remains at a reasonable average of 2.4475, indicating sufficient satellite geometry for feasible positioning. This low elevation distribution challenges the receiver’s ability to maintain lock, yet the stable PDOP and tracking results confirm that the system can still provide useful positioning data.

Figure 11.

Navigation solution at 12:31:37

The top plot shows stable variations in the East (E), North (N), and Up (U) coordinates over time. Despite the low SNR, the SDR receiver demonstrates consistent horizontal (E and N) positioning, with only minor fluctuations in the vertical (U) component. This stability suggests that the receiver successfully tracks IRNSS signals, achieving reliable positioning accuracy. The variations remain within acceptable limits, proving that even with weaker signals due to low elevation, the receiver can maintain accurate positional data.

The scatter plot shows a dense clustering of positional measurements around the mean position, with a horizontal spread of approximately ±10 meters. This tight clustering is indicative of reliable horizontal positioning accuracy, affirming that the SDR receiver can sustain accurate positioning using IRNSS signals, even in a low-elevation, low-SNR environment. This result proves the system’s capability to deliver consistent navigation performance suitable for general applications in Vietnam.

We found very interesting information about satellites PRN 5 and PRN 7 of the NavIC signal. Even though these satellites could be acquired perfectly as shown in Figure 8, there are no navigation messages in the navigation bit after the tracking phase. Figure 12 shows the tracking results of PRN 5 and PRN 7 and we can see a strange pattern of navigation message bits.

Figure 12.

Tracking plots for PRN 5 and PRN 7 on May 23rd 2024

Figure 13 illustrates the navigation bits after FEC decoding of four subframes (1168 bits in total, 292 bit for each subframe). The small red circles mark all the bit number 28th and 29th for every subframe-these two bits indicate the subframe id. It can be seen that PRNs 5 and 7 contain only subframe 4 and lacking all other subframes. This indicates that the operation of satellites 5 and 7 were not correct at the time of our experiment (they might be in the maintenance progress). Due to this reason, these two PRNs will be removed from the final PVT solution equations and thus reduce the accuracy of the final position solution.

Figure 13.

Navigation bits after FEC decoding of PRN 5 and PRN 7

Figure 14 shows IRNSS satellite acquisition results across several days and different times, illustrating the consistency in visible satellite PRNs and variations in acquisition metrics over time.

Figure 14.

Acquisition results for different days

Across all four subplots, the visible satellites (indicated by green and blue bars) remain consistent. This is evident as the same PRNs are observed in each acquisition attempt, confirming that the receiver detects a similar set of IRNSS satellites during each observation. This consistency suggests that the receiver is operating within a stable coverage area for IRNSS, with predictable satellite visibility patterns in this region.

The green bars indicate successfully acquired signals, whereas blue bars represent satellites that are visible but not acquired. While PRNs like 5, 7, and 10 are consistently acquired, others, such as PRN 2 and PRN 12, exhibit more instances where they are visible but not acquired, suggesting that these satellites may be at lower elevation angles or have weaker SNR, challenging acquisition stability.

Although the visible satellites remain the same, the acquisition metrics (height of the bars) fluctuate across different times and dates. For example, PRN 5 consistently shows high acquisition metrics in all instances, indicating a strong and reliable signal, while other satellites, such as PRN 1 and PRN 11, have lower acquisition metrics, reflecting weaker signals. This variation could be attributed to environmental factors, changes in atmospheric conditions, or fluctuations in signal strength due to satellite positioning.

The consistency in visible satellites combined with a stable acquisition of key satellites (e.g., PRNs 5, 7, and 10) demonstrates that IRNSS signals are feasible for use in Vietnam. Despite some fluctuations in acquisition metrics, the receiver can reliably acquire multiple satellites, providing a robust basis for positioning in this area.

The chart in Figure 15 displays the Carrier-to-Noise Density Ratio (C/N0) for 14 IRNSS satellites over a 24-hour period on June 1, 2024. The results highlight variations in signal strength across different satellites and time, reflecting the impact of satellite geometry and environmental factors on signal quality.

Figure 15.

CN0 for 24 hours on Jun 01st

Some satellites, such as PRNs 1, 5, and 7, maintain relatively high C/N0 values around 45–50 dB-Hz throughout the entire day. This stability indicates strong, reliable signals that are likely from satellites at higher elevation angles with consistent visibility. These high C/N0 values suggest that these satellites provide robust tracking capabilities, even over extended periods.

PRNs 3, 6, 8, and 10 display moderate C/N0 values, ranging between 40 and 45 dB-Hz, with some fluctuations over the 24-hour period. These satellites remain visible and provide usable signals, although the moderate C/N0 values suggest that they may be positioned at lower elevations or experience occasional attenuation due to environmental factors.

Notably, some satellites, such as PRNs 6 and 9, show abrupt drops in C/N0 to very low levels (near 0 dB-Hz) at specific times (around 3–5 hours and 17–19 hours). These dropouts indicate temporary loss of signal, possibly due to satellites moving below the horizon or entering regions of poor geometry. Such dropouts affect tracking continuity and may require additional satellites for stable positioning.

The presence of multiple satellites with stable and high C/N0 values suggests that IRNSS can be feasibly used for navigation in Vietnam over a 24-hour period. Satellites with high or moderate C/N0 ensure reliable signal availability for most of the day, while the occasional dropouts from lower C/N0 satellites may have minimal impact on overall positioning performance.

The skyplot (Figure 16) provides a visual representation of the IRNSS satellites’ positions over a 24-hour period, supplementing the Carrier-to-Noise Density Ratio (C/N0) chart above. The skyplot shows the trajectories of various satellites in terms of azimuth and elevation, which directly impact signal strength and acquisition quality observed in the C/N0 chart.

Figure 16.

Skyplot for 24 hours

Figure 17.

Skyplot for 48 hours

The plot reveals that the majority of the satellites are positioned at low elevation angles, particularly along the southwestern and southern horizons. PRNs 2, 6, and 9 appear to have extended trajectories near the horizon (elevations below 30 degrees). Low elevation angles generally result in lower C/N0 values due to increased atmospheric attenuation and multipath interference, explaining the weaker signal strengths observed in the C/N0 chart for these satellites.

Satellites such as PRN 3 and PRN 10 have relatively short paths, confined to low elevations in the southeastern part of the sky. This limited movement suggests that they are likely geostationary or geosynchronous satellites, which stay in a fixed position relative to the observer. Their positioning also leads to low-to-moderate C/N0 values, as seen in the previous chart, further supporting the conclusion that IRNSS signals in Vietnam are mostly received at low elevations.

Despite the low elevations, the skyplot confirms that certain satellites remain visible throughout the day, supporting continuous positioning. Although low-elevation angles generally reduce signal quality, the presence of multiple satellites, even at low elevations, proves that IRNSS can still be feasibly used in Vietnam. However, low C/N0 values may necessitate receiver optimizations or hybrid GNSS solutions to improve reliability.

Table 4 presents position errors (in meters) measured over 24-hour periods on three consecutive days (May 31st to June 3rd). Each entry shows the position error at specific timestamps, allowing a comparison of error fluctuations over time and across different days.

The position errors generally fall within a range of approximately 16–30 meters on most timestamps across all days, indicating reasonably stable positioning performance. This consistency in error magnitude suggests that the receiver can achieve feasible accuracy in this environment.

On certain timestamps, notably on June 2nd to June 3rd, there are significant spikes in position errors (e.g., 116.7896 meters at 15h08, 287.0283 meters at 16h08). These high errors could be due to temporary loss of signal, poor satellite geometry, or environmental factors such as signal obstruction or multipath effects, which compromise positioning accuracy.

Despite occasional high error spikes, the majority of position errors remain under 30 meters, which supports the feasibility of using IRNSS for positioning in Vietnam. While the system provides reasonably accurate results for most of the day, temporary errors indicate that additional measures, such as integrating with other GNSS or applying correction techniques, may enhance reliability for precise applications.