4.3 On-The-Fly Variance Estimation
4.4.1 Numerical Simulation
A numerical simulation is run in order to compare the performance attained by using different discriminators. The input signal (data and pilot channel) is generated with a constant frequency offset (ΞπΈπΈ) of 20 Hz, and an acceleration offset of 1 Hz/s2. The former is an arbitrary frequency offset to allow the FLL to gain lock, and the latter is to test the ability of the FLL to track a signal with maximum frequency rate as faced by a static GPS receiver (Tsui 2000). Four different FLL implementations are allowed to track the input signal simultaneously. The implementations differ by the choice of discriminator:
i. data channel only with π·π·πππΈπΈ, ii. pilot channel only with π·π·πππππππΏπΏπΏπΏ,
iii. standard discriminator combination with π·π·πππΈπΈ and π·π·πππππππΏπΏπΏπΏ, and
iv. Costas discriminator combination with π·π·πππΈπΈ on both data and pilot channels.
4.4.1.2 Test Methodology
The noise bandwidth of all the implementations is fixed to 4 Hz for fair comparisons and the loop filters designed in the analog domain are implemented using bilinear transformations. The predetection interval is set to 5 ms. Since this is a numerical simulation, instantaneous values of original input frequency and acceleration are known. These are compared with the frequency estimates used by the local oscillator in order to
measure the average frequency error of each predetection interval. A total of 3000 such frequency error values, measured in steady state, are then used to compute the frequency jitter due to thermal noise (ππππβπ π ππππππππ) and the steady-state error (ΞπΈπΈπΏπΏπΏπΏ). As defined by Kaplan (2006), frequency lock is declared if the frequency jitter (πππππ π ππ) satisfies the condition
3Οnet = 3ΟFLL + Ξfss β€4T1
coh (Hz) (4.23)
where the threshold of 1
4ππππππ β corresponds to the pull-in range of the cross product
discriminator with decision feedback (π·π·πππΈπΈ). Although the pull-in range of the cross product discriminator (π·π·πππππππΏπΏπΏπΏ) is twice as that of π·π·πππΈπΈ, this threshold is chosen based on the minimum pull-in range in the combination. For statistical reliability of the results, ππππβπ π ππππππππ and ΞπΈπΈπΏπΏπΏπΏ are averaged across 100 such independent runs of the FLL for each
considered C/N0. For each run, the FLL transient is allowed to settle, and the errors are
measured at steady state. Only those points that have a probability of frequency lock of at least 75% for a given C/N0 are used in the analysis. The frequency jitter is measured for a
C/N0 in the 15 to 40 dB-Hz range.
4.4.1.3 Analysis of Results
Figure 4-7 shows the simulation results. For C/N0 greater than 33 dB-Hz, the
performance of π·π·πππππππΏπΏπΏπΏ on the pilot channel and π·π·πππΈπΈ on the data channel are the same. Data bit predictions are more reliable under this situation; hence, there is no significant difference in performance. Under these C/N0 conditions, the standard and Costas
For C/N0 lower than 33 dB-Hz, increased noise is introduced by the data bit
decision process (Natali 1986). This is observed on the curve corresponding to the FLL on the data channel (π·π·πππΈπΈ). The Costas discriminator combination, which suffers from the same drawback, continues to provide nearly a 3 dB noise variance reduction with respect to π·π·πππΈπΈ. The standard discriminator combination shows a slightly degraded performance in comparison with the pilot channel only for low C/N0. The reason for this degradation
in performance is explained by analyzing the noise performance of π·π·πππΈπΈ used in the combination.
The π·π·πππΈπΈ relies on a differential data bit decision (πΏπΏππππ(β[. ])) to account for the effect of the data bit presence. The noise οΏ½ππ(ππ)οΏ½ introduced by the data bit decision part of the discriminator can be given by a model approximated to the first order as (Natali 1986):
Figure 4-7: Comparison of frequency jitter across different implementations to track data and pilot channels
Ο΅(t) = E[Ο΅(t)] + Ο΅1(t) = (1 β 2Pe) + Ο΅1(t) (4.24)
Pe=12 expοΏ½βπππΆπΆ
0ππππππβοΏ½ (4.25)
where πππ π is the probability of data bit error (Pe), πΈπΈ[ππ(ππ)] = (1 β 2πππ π ) accounts for the change in discriminator gain due to data bit errors, and ππ1(ππ) accounts for the noise introduced by faulty decisions. The autocorrelation function οΏ½π π ππ1(ππ)οΏ½ of ππ1(ππ) is given by (Natali 1986): π π ππ1(ππ) = β© βͺ β¨ βͺ β§2πππ π οΏ½[2(1 β πππ π )] β οΏ½ππππ ππππβοΏ½ οΏ½ 0 < ππ < ππππππβ 2πππ π (1 β 2πππ π ) οΏ½2 β οΏ½ππππ ππππβοΏ½οΏ½ ππππππβ < ππ < 2ππππππβ 0 ππππβπ π πππππ π πΏπΏπ π . (4.26)
Although the difference in noise variance at the discriminator output of π·π·πππΈπΈ and π·π·πππππππΏπΏπΏπΏ in a standard combination is accounted for by the weighting approach, the change in gain across π·π·πππΈπΈ and π·π·πππππππΏπΏπΏπΏ at lower C/N0, as given by πΈπΈ[ππ(ππ)] in Eq.(4.24), is unaccounted
for in the combination. Hence, a difference in performance is observed.
In terms of the minimum C/N0 required for tracking, as observed from Figure
4-7, the pilot channel has a frequency tracking threshold that is 6 dB lower than the data channel. The standard discriminator combination performs quite close (1 dB difference in minimum required C/N0) to the pilot-channel-only frequency tracking, whereas the
Costas discriminator combination performs close to the data-channel-only frequency tracking.
4.4.2 Validation of Results with Live Signals