Theoretical background

Theoretically, in bender element analysis, the arrival of a signal of certain frequencies is

determined based on the abrupt change in the amplitude, as well as the associated change in the
energy. In line with this concept, where significant energy is being measured upon the arrival of
the S-waves, the exact amplitude of a signal at any time 𝑡 and frequency 𝑓 has to be fully returned
upon any transformation to avoid any biases and discrepancies in the analysis. Tackling this
problem, the Stockwell Transform (ST) based method, which is originally introduced to deal with
the non-stationary signals in the geophysical data (Stockwell et al. 1996; Stockwell 1999), is
applied in this study. For the sake of brevity, only the main features of the ST based method that
are assisting in the arrival-time determination are highlighted herein, and comprehensive reviews
of this method could be found in (references).

The generalized Stockwell Transform of a signal ℎ(𝑡) is defined as:

∞
𝑆(𝜏, 𝑓, 𝑝) = ∫ ℎ(𝑡)𝜔(𝜏 − 𝑡, 𝑓, 𝑝) 𝑒 −𝑗2𝜋𝑓𝑡 𝑑𝑡 (1)
−∞

where 𝑡 denotes the time, 𝜏 denotes the time of spectral localization, 𝑓 denotes the frequency of
spectral localization, 𝑝 denotes a set of parameters which determine the shape and properties of
the window function 𝜔. In this study, the Slepian window is selected as the window function. By
maximizing the energy concentration in the main lobe, it works excellent in averaging out noises
in the spectrum and reducing information loss at the edges of the window, thereby precisely
localizing the responses in both time and frequency domains (references). Comparing to other
common methods in signal processing, e.g., Wavelet Transform (WT) and Short-Time Fourier
Transform (STFT) methods, the ST based method shows outstanding properties, which is vital in
the arrival-time determination. First, using a variable window function that changes in width, the
ST based method provides a progressive resolution in both time and frequency domains, unlike
the STFT method where a tradeoff between the time and frequency resolution is unavoidable due
to the fixed window width. Second, the ST based method features the absolutely referenced phase
information, where the information transformed is always referenced to the time-origin, which is
not available from locally reference phase obtained by the WT method. Third, the ST based method
attains the frequency invariant amplitude response, where the amplitude is fully returned regardless
of the frequency, in contrast to the WT method where the amplitude attenuates at high frequencies.
Combining all these promising characteristics, the ST based method accurately transforms the
signals recorded into a time-frequency representation, of which any points in this plane represents
the energy (amplitude) at 𝑡 and 𝑓. With the high-resolution time and frequency axes and precise
referencing of the exact amplitude to the time-origin, this aids the users in precisely, yet objectively
localizing the arrival-time of S-waves, and details of this implementation is discussed in the next
part.

Numerical Experiment

A simulated synthetic signal is first used to validate the reliability and workability of the ST based
method on the arrival-time determination in bender element tests. Guided by the procedure adopted
by Sánchez-Salinero et al. (1986) and Wang et al. (2007), the numerical experiment in this study
is modified to accommodate the parameters used in the physical experiment. The main steps
involved and parameters selected are delineated as follows, and detailed explanation could be
found in Wang et al. (2007).

During the wave generation, propagation and receiving processes, the transfer functions,
which may change or distort the input signal, 𝑆𝑖𝑛𝑝𝑢𝑡 , have to be considered in obtaining the
appropriate the output signal, 𝑆𝑜𝑢𝑡𝑝𝑢𝑡 . Generally, the expression is given as

𝑆𝑜𝑢𝑡𝑝𝑢𝑡 = 𝑆𝑖𝑛𝑝𝑢𝑡 ∙ 𝐻𝑆 (ω) ∙ 𝐻𝑠𝑜𝑖𝑙 (ω) ∙ 𝐻𝑅 (ω) (X)

where 𝐻𝑆 (ω) is the transfer function when source bender element generating signal; 𝐻𝑠𝑜𝑖𝑙 (ω) is
the transfer function when wave propagating in the soil sample; and 𝐻𝑅 (ω) is the transfer function
when receiver bender element receiving signal. Among the transfer functions, 𝐻𝑆 only creates a
small delay in the output signal, which could be neglected from the practical point of view,
according to the results of the experimental test done by Wang et al. (2007); therefore it is not
considered in this study. Next, in assessing the near-field effects, the following 𝐻𝑠𝑜𝑖𝑙 , which is
derived based on the analytical solution for an infinite isotropic elastic medium proposed by Cruse
and Rizzo (1968), as well as adopted by (references), is used:

(Equations used)

where XXX. The values of the parameters are similar to those used in the physical experiment in
this study to provide comparable results, as listed in the Table X. Similar to the previous researches
of bender element tests (e.g., Jovičić et al. 1996; Lee and Santamarina 2005; Wang et al. 2007), it
is observed that 𝐻𝑠𝑜𝑖𝑙 seemingly creates only a near-field effect in front of the input sine pulse, as
shown in Fig. X. Then, the transfer function of the receiver bender element, 𝐻𝑅 , is estimated based
on the response of a damped single-degree-of-freedom system subject to a pulse. According to
Santamarina and Fratta (1998), this time-domain response function is given as:

𝑒 −𝐷𝜔𝑛 𝑡 2
ℎ𝑅 (𝑡) = 𝑠𝑖𝑛 (𝜔𝑛 √1 − 𝐷𝑏𝑒 𝑡) (X)
2
𝑚𝜔𝑛 √1−𝐷𝑏𝑒

where 𝜔𝑛 = natural angular frequency; 𝐷 = wave attenuation coefficient; 𝐷𝑏𝑒 = damping ratio of
the bender element system; and m = system mass; all these values selected are listed in the Table
X. Then, the 𝐻𝑅 is obtained from the fast Fourier transform of the ℎ𝑅 , i.e., 𝐻𝑅 (𝜔) = FFT[ℎ𝑅 (𝑡)].

Summarizing, the computation steps used in the numerical experiment are shown as
follows (similar to Lo 2005 and Wang et al. 2007):

1. Define an input excitation (single sine pulse used in this study): 𝑥(𝑡). The frequency used is X
kHz.

2. Compute the Fourier transform of the input signal: 𝑋(𝜔) = FFT[𝑥(𝑡)]. The time interval used
is X s, and the sampling points in the time domain, N = x are used in the fast Fourier transform.

3. Select proper parameters for the analytical transfer function 𝐻𝑠𝑜𝑖𝑙 (refer to Table X for the
values of the parameters selected, and Eq (X) for the expression of 𝐻𝑠𝑜𝑖𝑙 ).

4. Define the resonant response for the receiver bender element (damped single-degree-of-
freedom system) in the time domain: ℎ𝑅 (𝑡). (refer to Table X for the values of the parameters
selected, and Eq (X) for the expression of ℎ𝑅 ).

5. Compute the Fourier transform of this resonant response: 𝐻𝑅 (𝜔) = FFT[ℎ𝑅 (𝑡)].

6. Compute the inverse Fourier transform to obtained the simulated time-domain output 𝑦(𝑡):
𝑦(𝑡) = IFFT{FFT[𝑥(𝑡)] ∙ 𝐻𝑠𝑜𝑖𝑙 (𝜔) ∙ 𝐻𝑅 (𝜔)}.
Fig. 1b presents the simulated output signal 𝑦(𝑡) of the input signal 𝑥(𝑡) due to both transfer
functions 𝐻𝑠𝑜𝑖𝑙 and 𝐻𝑅 . To recover the actual situation happening in the physical experiment as
much as possible, random noises, which act as the ambient noises in the physical experiment, is
combined with the simulated output signal, as shown in Fig. 1c.

Subsequently, using the ST based method, the combined simulated output signal is
transformed into the time-frequency spectrum, as shown in Fig. 2a, where the shade of color
represents the energy density. To implement this transformation, across the time domain, the
frequency between 0 – 100 MHz is first divided into 1 × 104 parts, i.e. the window width along the
frequency is fixed constant as 100 Hz. Then, at each frequency part, the energy concentration in
the main lobe of the window function (i.e., Slepian window used in this study) is maximized,
which accordingly varies the window width along the time domain, to reproduce high-resolution
responses in the time-frequency representation. Afterwards, a filter is applied along the frequency
domains to obtain the regions where the energy density is significant, i.e., 8 – 12 kHz. Instead of
transforming the results back to time-domain for determining the arrival-time of S-waves, which
is demonstrated by Wang (2006), a more objective alternative is used in this study to avoid the
highly subjective manual arrival-time determination at the time-domain. As described previously,
the arrival of S-waves is associated with a significant change in the energy measured; this is used
as the objective criterion to determine the arrival-time in this study. Therefore, as shown in Fig.
2b, at each time t, the energy in the range of 8-12 kHz (i.e., the range with significant energy
observed) is summed up, followed by the computation of energy gradient, as shown in Fig. 2c.
The maximum energy gradient point (see the cross symbol in Fig. 2c) is chosen as the arrival-time
of the S-waves. Comparing to the ground-truth, i.e., X ms (see Fig. X), the time estimated based
on the ST based method is X ms, which is acceptable yet is obtained using an unbiased and
objective approach.

References:

Y.-H. Wang "The Tutorial : S-transform " National Taiwan University Taipei Taiwan ROC Taiwan
2006.

Using the ST based method, the time-frequency spectrum of the signal shown in Fig. 1c is
computed, as shown in the Fig. 2a, where the shade of color represents the energy density. Across
the time domain, the frequency between 0 – 100 MHz is divided into 1 × 104 parts

As mentioned, to localize the responses at high-resolution for both time and frequency domains,