Imp mprovin ing S Sensitivity o on K Kea CubeS eSat GPS GPS Rec ecei eiver ers Eamonn Glennon
Kea CubeSat GPS Receivers • Follow-on from Namuru V3.2R3A developed for DST-Group Biarri program • L1 C/A code GPS receiver with features designed to support in-orbit and high dynamics operation • SmartFusion 2 SoC & BL2627 RFFE • To be flown on 3 CubeSat missions – Buccaneer Risk Mitigation – UNSW QB50 EC0 and – USyd QB50 iInspire 2 2 IGNSS 2016, 6-8 Dec 2016, UNSW,Sydney
High Sensitivity in Space • Some space applications have weak signals • GPS Radio Occultation – Signals passing through the atmosphere’s limb are weak and subject to ionospheric fading – Continuous phase lock needed for carrier phase observations • GPS in medium, high or geostationary orbit – Space Service Volume aims for consistent performance in MEO, HEO & GEO – Navigation requires use edges of GPS beam pattern and side- lobes | 3 IGNSS 2016, 6-8 Dec 2016, UNSW,Sydney
Space Service Volume http://www. gps .gov/governance/advisory/meetings/2015-06/bauer.pdf | 4 IGNSS 2016, 6-8 Dec 2016, UNSW,Sydney
High Sensitivity GPS • High sensitivity techniques developed for cell-phone AGPS – Combination of longer coherent integration (CI) periods (i.e. separate accumulation of in-phase and quadrature phase channels) – Up to 20 ms without data wiping – Longer CI reduces squaring losses, but increases number of searches – Non-coherent integration (NCI) for additional sensitivity (i.e. accumulation of magnitude or magnitude squares of I & Q) • Acquisition assistance for reduced search times – Less feasible for MEO, HEO & GEO, but TLEs could be used • Search acceleration hardware – Useful for acquisition, but not for tracking | 5 IGNSS 2016, 6-8 Dec 2016, UNSW,Sydney
KeaV41SBR4 Modifications • Pre modification sensitivity not great! – 4 ms coherent integration (CI), so bandwidth is 250 Hz – 8 non-coherent rounds of integration (NCRI) – PLL ran at 2 x 4 ms, but detection as above • Improvements required – Ability to support up to 20 ms CI, with 4+ NCRI – DSP produces ½ dumps for FLL to allow FLL to run at the same rate as the PLL – PLL runs at up to 20 ms, so bandwidth is 50 Hz – Detection using 20 ms x 4+ NCRI – Automatic switch to higher CI when bit-synchronisation achieved | 6 IGNSS 2016, 6-8 Dec 2016, UNSW,Sydney
Sensitivity Predictions Predicted Sensitivity for Different Coherent/Non-coherent Integrations 35 30 25 4 ms x 8 NCR : SNR (dB) Output SNR (dB) 20 10 ms x 4 NCR : SNR (dB) 15 10 ms x 8 NCR : SNR (dB) 8.4 dB 6.8 dB 10 20 ms x 4 NCR : SNR (dB) 5 22 27 32 37 Input C/N0 ( dBHz) van Diggelen Sensitivity Spreadsheets | 7 IGNSS 2016, 6-8 Dec 2016, UNSW,Sydney
Adaptive Thresholds • High sensitivity requires SNR to be maximized • Longer coherent integration reduces squaring losses • Additional non-coherent integration further improves SNR BUT • Detection requires comparison of CI & NCI against a detection threshold • Question: How to set the detection threshold? • Answer: Adaptively estimate the threshold using a correlator noise finger/tap T = μ + k σ μ & σ are the (estimated or measured) mean & standard deviation of the NCIs k is a constant that sets the false alarm rate (FAR) | 8 IGNSS 2016, 6-8 Dec 2016, UNSW,Sydney
Measure σ 1ms Calculate σ 1ms statistic from 1 ms μ NCI (σ 1ms ) & σ NCI (σ 1ms ) I & Q dumps assuming NC & NCR CI & NCI Selected Coherent integration (CI) & Threshold Non coherent integration (NCI) T = μ NCI (σ 1ms ) + K σ NCI (σ 1ms ) σ 1ms values update at 1 kHz, so faster response time • • Not effective in the presence of Multiple Access Interference (MAI) / Cross Correlation Noise | 9 IGNSS 2016, 6-8 Dec 2016, UNSW,Sydney
Noise Tap Usage – NC & NCI statistics/histogram Noise Calculate ΣN & ΣN 2 Noise tap N processed Threshold μ NCI samples from which T c = μ NCI + K σ NCI exactly as E, P, L taps. σ NCI μ NCI & σ NCI are obtained Construct histogram of Noise samples Threshold T h • Slow update rate of 1000/(CI*NCR) Hz / input sample • Need (say) 100 samples for meaningful statistics Threshold calculated from μ NCI & σ NCI assumes Normal distribution • – BUT this is not necessarily true (approaches Normal ~ Central Limit Theorem) • Histogram method allows Th to be directly extracted with a given False Alarm Rate and can be maintained efficiently – See the paper for details | 10 IGNSS 2016, 6-8 Dec 2016, UNSW,Sydney
Matlab Experiment • Create signals at IF in Matlab that are down-converted and de-spread with noise & E,P,L taps • Four satellite scenarios considered – 1 strong, rest weak – 6 strong – 6 strong, SV1 affected by 1 SV MAI – 6 strong, SV1 affected by 2 SVs of MAI • Two different tracking scenarios – 4 ms & 8 NCR, same as legacy Namuru & Kea – 20 ms & 4 NCR, highest sensitivity setting Estimate μ NCI , σ NCI analytically, derived from 1 ms dump • statistic σ 1ms and from NCI noise samples directly | 11 IGNSS 2016, 6-8 Dec 2016, UNSW,Sydney
Matlab Experiment - Statistics Mean NC Noise for 4 ms x 8 NCR Mean NC Noise for 20 ms x 4 NCR 25 20 Mean NC Noise 20 15 15 Mean NC Noise 10 10 Mean(Est) Mean(Est) 5 5 Mean(Meas 1ms) Mean(Meas 1ms) 0 0 1 6 6 6 1 6 6 6 Meas Mean Meas Mean Strong Strong Strong, Strong, Strong Strong Strong, Strong, 1 MAI 2 MAI 1 MAI 2 MAI Test Scenario Test Scenario StdDev NC Noise for 4 ms x 8 NCR StdDev NC Noise for 20 ms x 4 NCR 7 7 StdDev NC Noise 6 StdDev NC Noise 6 5 5 4 4 3 3 2 StdDev(Est) StdDev(Est) 2 1 1 StdDev(Meas 1ms) StdDev(Meas 1ms) 0 0 Meas StdDev Meas StdDev 1 6 6 6 1 6 6 6 Strong Strong Strong, Strong, Strong Strong Strong, Strong, 1 MAI 2 MAI 1 MAI 2 MAI Scenario Scenario | 12 IGNSS 2016, 6-8 Dec 2016, UNSW,Sydney
Matlab Experiment - Thresholds Thresholds for 4 ms x 8 NCR Using Various Techniques 50 T(2 s1ms) Threshold Value T(3 s1ms) 40 T(2 s1ms) 30 T(3 s1ms) 20 T(2 sNC) T(3 sNC) 10 T5% 0 T1% 1 Strong 6 Strong 6 Strong, 1 MAI 6 Strong, 2 MAI Scenarios Thresholds for 20 ms x 4 NCR Using Various Techniques T(2 s1ms) 40 T(3 s1ms) 30 T(2 s1ms) Threshold T(3 s1ms) 20 T(2 sNC) 10 T(3 sNC) T5% 0 T1% 1 Strong 6 Strong 6 Strong, 1 MAI 6 Strong, 2 MAI Scenarios | 13 IGNSS 2016, 6-8 Dec 2016, UNSW,Sydney
Simulator Tests • Legacy firmware tested against modified firmware with Spirent GSS8000 power gradually reduced using a splitter – Improvement of 6 dB achieved (as predicted) BUT – Legacy firmware stopped navigating at -2 dB, but only lost lock at -1 dB, whereas new firmware navigated down to -4 dB – Removing signal level quality thresholds allowed legacy receiver to track & navigate down to -1 dB, so only a 3 dB improvement • Tuning the new receiver to use 10 ms x 8 NCR and repeating the test without the splitter gave the new firmware a 5 dB improvement over the legacy firmware • Once 20 ms x 4 NCR tracking is improved, an additional 1.6 dB can be expected. | 14 IGNSS 2016, 6-8 Dec 2016, UNSW,Sydney
Conclusion • Improved signal processing and other changes have improved Kea sensitivity by about 8 dB • Thresholds are now determined adaptively using corrrelator noise taps • Longer coherent integration reduces squaring losses • Further 1.5 dB improvement possible | 15 IGNSS 2016, 6-8 Dec 2016, UNSW,Sydney
Questions | 16 IGNSS 2016, 6-8 Dec 2016, UNSW,Sydney
Recommend
More recommend
