Galileo Masterclass Brazil (GMB) 2022

Lecture 2 - Spread Spectrum Ranging

Felix Antreich
Outline

Time-Delay Estimation

Signal Properties

Examples for Signal Design

1 / 23
Maximum Likelihood Time-Delay Estimation (1)
For
x = x[0]
let us assume a random variable x has a multivariate Gaussian
probability density function (pdf) parameterized by the
parameter τ and thus we get
" √ #
1 kx − P c(τ )k22
px (x; τ ) = exp −
(2πσn2 )N/2 2σn2

The likelihood function with respect to the parameter τ is given

as
L(x; τ ) = px (x; τ )

I L(x; τ ) is a function of the parameter τ , which is to be

estimated at a given realization of the random variable x
I The pdf px (x; τ ) is a function of the realization of the
random variable x for a fixed value of the parameter τ
2 / 23
Maximum Likelihood Time-Delay Estimation (2)
Now the maximum likelihood estimator (MLE) can be given as

τ̂ = arg max {L(x; τ )} = arg max {log (L(x; τ ))} .

τ τ

The MLE is asymptotically (large N) unbiased and efficient.

When further deriving the estimator we get

τ̂ = arg max {log (L(x; τ ))}

τ
   1 √ 
= arg max log(1) − log (2πσn2 )N/2 − 2 kx − Pc(τ )k22
τ 2σn
n √ o
= arg max −kxk22 + 2 P xT c(τ ) − P kc(τ )k22
τ

As the first term does not depend on τ and the third term is
constant√with kc(τ )k22 ≈ N, ∀τ as well as dropping the constant
factor 2 P we can write

τ̂ = arg max xT c(τ ) = arg max {J(τ )}


τ τ

3 / 23
Time-Delay Estimation with a Delay Locked Loop
(DLL) (1)
In practice, time-delay estimation is performed using a DLL
applying a gradient ascent method (step size µ > 0) where the
k th iteration can be given as
∂J(τ̂ [k − 1])
τ̂ [k] = τ̂ [k − 1] + µ
∂τ
The derivative can be approximated using the central difference
quotient of length 2∆,
µ
τ̂ [k] = τ̂ [k − 1] + (J(τ̂ [k − 1] + ∆) − J(τ̂ [k − 1] − ∆))
2∆
µ
xT c(τ̂ [k − 1] + ∆) − xT c(τ̂ [k − 1] − ∆)


= τ̂ [k − 1] +
2∆
A stochastic version (considering successive periods k) can be
given as
µ
τ̂ [k] = τ̂ [k − 1] + xT [k − 1]c(τ̂ [k − 1] + ∆)
2∆
−xT [k − 1]c(τ̂ [k − 1] − ∆)

4 / 23
Time-Delay Estimation with a Delay Locked Loop
(DLL) (2)
Without noise, assuming that the receiver uses signal matched
correlators, and the time-delay tracking error
ε = τ − τ̂
the discriminator S-curve for a coherent early-late DLL is
S(ε) = Rc (ε − ∆) − Rc (ε + ∆)
S[k ; ε[k]] = Rc [ε[k] − ∆] − Rc [ε[k ] + ∆].

Rc (ε)

k
early late

k −1
late
early

ε 5 / 23
Cramer-Rao Lower Bound (CRLB) (1)
If  
∂ log (L(x; τ ))
E =0
∂τ
the variance of the time-delay estimation error στ2 of any
unbiased estimator is lower bounded by the Cramer Rao lower
bound (CRLB)
σn2
1
var(τ̂ ) = στ̂2 ≥ h i= P
∂cT (τ ) ∂c(τ )
∂ 2 log(L(x;τ ))
−E ∂τ 2 ∂τ ∂τ

which leads to
Bn 1
στ̂2 ≥ P
R∞
2 |P(f )|2 df
2
4π −∞ f
N0

where Bn is the equivalent noise bandwidth of the generic

estimator and Z ∞
|P(f )|2 df = 1
−∞
6 / 23
Cramer-Rao Lower Bound (CRLB) (2)
R∞
The term −∞ f 2 |P(f )|2 df is:
I Second moment of the power spectrum
I Root mean square (RMS) bandwidth
I Gabor bandwidth
It is equal to the curvature of Rc (ε) at ε = 0
∞
1 d 2 Rc (ε) 
Z 
2 2
f |P(f )| df = − 2
−∞ 4π d ε2 ε=0

and if p(t) is band-limited to [−B, B] it is upper bounded by

Z B Z B Z B
2 2 2 2 2
f |P(f )| df ≤ B |P(f )| df ≤ B s.t. |P(f )|2 df = 1
−B −B −B

Thus, |P(f )|2 = 12 (δ(f − B) + δ(f + B)) maximizes the Gabor

bandwidth and p(t) = cos(2πBt) or p(t) = sin(2πBt).
7 / 23
Outline

Time-Delay Estimation

Signal Properties

Examples for Signal Design

8 / 23
Synchronization Accuracy
I The higher the Gabor bandwidth of the signal, the higher
the synchronization accuracy that can be achieved in
terms of the CRLB
I The second moment of the power spectrum of a signal with
bandwidth B is upper bounded by B 2 ⇒ The higher the
available signal bandwidth , the higher the possible
synchronization accuracy
I A high processing gain G is desireable (large bandwidth B)
⇒ high synchronization accuracy, high interference
robustness, and low MAI-A
I Minimizing the CRLB for τ and maximizing time
concentration are contradictive tasks

We can state:
The lower the CRLB for τ (CRLB → B12 ) ⇒ the lower time
concentration of p(t) (p(t) → sin(2Bπt) or cos(2Bπt))
9 / 23
Time Concentration (1)
Time and frequency concentration of p(t) can be given by the
quantities:
R Tc /2 2 RB
−Tc /2 |p(t)| dt |P(f )|2 df
α= R∞
2
β = R −B
∞ 2
−∞ |p(t)| dt −∞ |P(f )| df

with Z ∞ Z ∞
2
|P(f )| df = |p(t)|2 dt = 1
−∞ −∞

and p(t) being strictly band-limited to [−B, B] ⇒ β = 1, and

(uncertainty principle of Fourier transform) we get
Z Tc /2
|p(t)|2 dt = α < 1
−Tc /2

10 / 23
Time Concentration (2)

The maximum of the absolute value of the sidelobes of Rc (ε) is

given as
∀ |νi | ≤ κ, κ ∈ [0, 1]
i∈N

νi denote the value of Rc (ε) at the sidelobes, besides the global

maximum of Rc (ε) at ε = 0.

We can state:
1. The higher the sidelobes of Rc (ε) ⇒ the higher the
sidelobes of p(t) ⇒ the lower time concentration of p(t)
2. The higher the sidelobes of Rc (ε) ⇒ the higher κ ⇒ the
less robust the estimation of τ (likelihood has local maxima
besides the global maximum)

11 / 23
Multiple Access Interference (1)
I MAI-A and MAI-R can be considered as additional
interference components with zero mean
I In general both MAI-A and MAI-R are dependent on the
propagation characteristics of the transmitted signal
I U users (e.g. visible GNSS satellites) with u = 1, . . . , U
and power Pu causing MAI-A
I V users of another system (e.g. visible satellites of a
different GNSS) with v = 1, . . . , V and power Pv are
causing MAI-R
I The received signal of another system (e.g. different
GNSS) in the same frequency band has PSD ΦR (f )
I The reference PR sequence generator is perfectly
synchronized with the received desired signal with power
P, so the time-delay τ of the desired signal is known
I The receiver was able to perform down conversion,
matched filtering with P ∗ (f ), and sampling at the chip
duration
12 / 23
Multiple Access Interference (2)
We can define the statistics of the matched filter output for a
WSCS sequence {dm } ∈ {−1, 1} with period Td = Nd Tc as
P
SINR =
PN + PA + PR
where the noise power can be given as
Z ∞
1 N0 N0 N0
PN = |P(f )|2 df = =
Nd Tc 2 −∞ 2Nd Tc 2Td
the power of the MAI-A can be given as
U Z ∞
1 X
PA = Pu |P(f )|4 df
2Nd Tc −∞
u=1

and the power of the MAI-R can be given as

V Z ∞
1 X
PR = Pv |P(f )|2 ΦR (f ) df
2Nd Tc −∞
v =1

13 / 23
Multiple Access Interference (3)
I The background noise is assumed as white Gaussian
noise with spectral density of N0 /2
I The variance due to background noise can be considered
as white noise of density N0 /2 filtered by the transfer
function of the receive filter P ∗ (f )
I All u users (e.g. visible GNSS satellites) are assumed to be
independent and unsynchronized with the desired signal
I It is assumed that their time-delays are independently
uniformly distributed in [0, Tc ] and their phases are
independently uniformly distributed in [0, 2π]
I The effect of the u-th user (e.g. GNSS satellite) on the
matched filter output of the desired signal will be that of
white noise passed through the tandem combination of two
filters with the transfer functions |P(f )|2 ⇒ MAI-A
I Similar assumptions as for the u users above can be taken
for the v users ⇒ MAI-R
14 / 23
Multiple Access Interference (4)
The ratio of the SNR to the SINR can be given as

SNR P/PN P + PR
∆SNR = = =1+ A
SINR P/(PN + PA + PR ) PN
U V ∞
Pu B
Z Z
X X Pv
= 1+ |P(f )|4 df + |P(f )|2 ΦR (f )df
N0 −B N0 −∞
u=1 v =1

The variance of the MAI-A component can be lower bounded by

Z B
1
|P(f )|4 df ≥
−B 2B

with equality iff

1 √ sin(2πBt)
|P(f )|2 = , −B ≤ f ≤ B, p(t) = 2B
2B 2πBt

15 / 23
Multiple Access Interference (5)
A proof can be given by using the Schwarz inequality:
"Z #2
Z B Z B B
2 2
|X1 (f )| df |X2 (f )| df ≥ X1 (f )X2 (f ) df
−B −B −B

RB
Suppose that X1 (f ) = |P(f )|2 , X2 (f ) = 1 and −B |P(f )|2 df = 1,
it follows that
"Z #2
Z B B
|P(f )|4 df · 2 B ≥ |P(f )|2 df =1
−B −B

Thus we get,
Z B
1
|P(f )|4 df ≥
−B 2B

16 / 23
Multiple Access Interference (6)

I Assumptions for MAI-A and MAI-R are derived following

IS-95 and CDMA2000 standards
I In GNSS MAI-A and MAI-R are called spectral separation
R∞
I The term −∞ |P(f )|2 ΦR (f )df is called spectral separation
coefficient (SSC)
R∞
I The term −∞ |P(f )|4 df is called self SSC
I MAI-A and MAI-R has to be considered in signal design
(chip pulse shape design) or can be treated in the receiver
(multi-user detection and mitigation)
We define the CRLB-I as the CRLB which considers noise plus
interference (MAI-A, MAI-R):

σ̃τ̂2 ≥ στ̂2 · ∆SNR

17 / 23
Outline

Time-Delay Estimation

Signal Properties

Examples for Signal Design

18 / 23
Example Pulse Shapes
In order to illustrate the previously discussed signal or pulse
shape properties we consider:
I Three example pulse shapes p(t) with κ = 0.1, κ = 0.5,
κ = 0.7, B = 1.023 MHz, and BTc = 1
I Pu = −154 dBW with U = 11
I N0 = −204 dBW/Hz
I c denotes the speed of light
I For this example we assume that no MAI-R is present
I The multipath error envelope gives the maximum bias of a
DLL in case that in addition to the line-of-sight signal a
single reflective multipath signal with signal-to-multipath
ration of 6 dB is present
I The envelope is defined by the cases if the multipath signal
has a relative phase of 0 or of π with respect to the
line-of-sight signal
19 / 23
Time and Frequency Domain

1.1
1 1 κ = 0.1
κ = 0.1 κ = 0.5

Normalized |P(f )|2

0.8 κ = 0.5 κ = 0.7
Normalized p(t)

κ = 0.7 0.8
0.6

0.4
0.6
0.2

0 0.4

−0.2
0.2
−0.4

−0.6 0
−3 −2 −1 0 1 2 3 −1.5 −1 −0.5 0 0.5 1 1.5

t/Tc f /B

20 / 23
Autocorrelation and CRLB-I

1
κ = 0.1 κ = 0.1
0.8
κ = 0.5 κ = 0.5
0.6 κ = 0.7 κ = 0.7

0.4

σ̃τ̂ · c
Rc (ε)

0.2

−0.2

−0.4 0
10
−0.6

−0.8 30 32 34 36 38 40 42 44 46 48 50
−3 −2 −1 0 1 2 3
P
ε/Tc N0 / dB-Hz

p(t) κ = 0.1 κ = 0.5 κ = 0.7 Lower bound MAI-A

MAI-A 0.79 0.80 0.99 0.54
∆SNR 1.79 (2.53 dB) 1.80 (2.55 dB) 1.99 (3.0 dB) 1.54 (1.87 dB)

21 / 23
Multipath Error Envelope and S-Curve

0.2 0.4
κ = 0.1
0.15 κ = 0.1 0.3
κ = 0.5
κ = 0.5
κ = 0.7
0.1 κ = 0.7 0.2
Bias in Tc

0.05 0.1

S(ε)
0 0

−0.05 −0.1

−0.1 −0.2

−0.15 −0.3

−0.2 −0.4
0 0.5 1 1.5 2 2.5 3 −2 −1.5 −1 −0.5 0 0.5 1 1.5 2

Relative delay of multipath in Tc ε/Tc

22 / 23
Summary

Signal properties for time-delay estimation and GNSS

1. The higher the Gabor bandwidth of the signal ⇒ the higher the synchronization
accuracy that can be achieved in terms of the CRLB
2. Minimizing the CRLB for τ and maximizing time concentration are contradictive
tasks
3. The higher the processing gain G (large bandwidth B) ⇒ the higher
synchronization accuracy and the higher interference robustness
1
4. The lower the CRLB for τ (CRLB → B2
) ⇒ the lower time concentration of p(t)
(p(t) → sin(2Bπt) or cos(2Bπt))
5. The higher the sidelobes of Rc (ε) ⇒ the higher the sidelobes of p(t) ⇒ the
lower time concentration of p(t)
6. The higher the sidelobes of Rc (ε) ⇒ the higher κ ⇒ the less robust estimation
of τ (likelihood has local maxima besides the global maximum)
7. MAI-A and MAI-R have to be considered in the signal design (chip pulse shape
design) or can be treated in the receiver (multi-user detection and mitigation)

23 / 23