Post-processing of background noise from SCPT auto source signal: A feasibility study for soil type classification Sung-Woo Moon a , Robin E. Kim b , Arthur C.C. Cheng c , Yunyue Elita Li c , Taeseo Ku c,⇑ a Department of Civil and Environmental Engineering, Nazarbayev University, 53 Kabanbay Batyr Ave, Nur-Sultan 010000, Kazakhstan b Department of Civil and Environmental Engineering, Hanyang University, Wangsimri-ro 222, Sungdong-gu, Seoul 15588, South Korea c Department of Civil and Environmental Engineering, National University of Singapore, 1 Engineering Drive 2, Singapore 117576, Singapore article info Article history: Received 3 September 2019 Received in revised form 12 January 2020 Accepted 10 February 2020 Available online 13 February 2020 Keywords: Seismic piezocone test (SCPTu) Background noise Cross power spectral density Cone tip resistance Sleeve friction abstract The continuous-interval seismic piezocone test (CiSCPTu) system has been recently developed for obtain- ing detailed information on shear wave velocity (V s ) along the depth. In the seismic signal processing, background noises are generally unfavorable since it can mask main shear waves. However, this study introduces a novel idea to utilize the unwanted noises to characterize soil types. The method is based on the integrated and normalized cross power spectral density ( b C y k ) of the background noises, calculated using a chosen reference signal, and the cone measurements (i.e., cone tip resistance (q t ) and sleeve fric- tion (f s )). We used the background noise recordings collected at Richmond (Canada) with well-defined geological layers. The developed correlations between background noise and cone measurements are ver- ified via two traditional CPT-based Soil Behavior Type (SBT) classification systems. The results show that the simplified soil types into sand and clay in SBT system predicted by the developed relationships (i.e., 92 to 95% of sand and 60 to 88% of clay) are comparable with the identification (i.e., 96 to 97% of sand and 69 to 99% of clay) with CPT measurements. Ó 2020 Elsevier Ltd. All rights reserved. 1. Introduction Since the in-situ shear wave velocity (V s ) is a critical parameter for determining the initial shear stiffness (G 0 or G max ), required for dealing with the geotechnical engineering problems such as static and dynamic deformations and liquefaction potentials, various geophysical methods have been utilized to obtain the in-situ V s . For example, invasive techniques such as downhole test (DHT) and crosshole test (CHT), and non-invasive techniques such as Spectral Analysis of Surface Waves (SASW) method [1] and Multi-Channel Analysis of Surface Waves (MASW) method [2] have been predominantly employed in exploration geophysics as well as geotechnical engineering applications. The conventional DHT and CHT tests are much more costly and time-consuming than the non-invasive techniques, albeit they pro- vide reliable in-situ V s . Alternatively, direct push downhole-type methods such as seismic cone penetration test (SCPT) and seismic flat dilatometer test (SDMT) can produce equivalent promising results for determining in-situ V s profiles [3–5]. Especially, the SCPT may be the most cost-efficient method and became an attrac- tive option when comparing typical costs of several geophysical tests [6]. Fig. 1(a) and (b) present the general setup and procedures of SCPT or SDMT to measure in-situ V s , including two types of com- mon approaches such as a pseudo-interval method using a single receiver (Fig. 1(a)), and a true interval method using multiple receivers (Fig. 1(b)). While a seismic peizocone is pushed into the ground, in-situ V s generated from a horizontal seismic source is measured at a single or dual seismic receiver generally per every 1 m-depth interval. In addition, cone tip resistance (q t ), sleeve fric- tion (f s ), and pore-water pressure (u 2 ) with depth are provided together from a single sounding. Based on the concept of an automatic mechanical gear system, a new portable automated seismic triggering source (i.e., RotoAuto- Seis), developed by McGillivray and Mayne [7], allows measuring the continuous variation of shear wave velocity (V s ) with depth, as well as typical readings of piezocone penetration test (CPTu) (i.e., q t , f s , u 2 ) during non-stopping cone penetration. Fig. 1(c) illus- trates the automated impulse source that can generate continuous shear waves consistently and repeatedly at a constant rate of speed (e.g., every 1 to 10 s). The schematic diagram of the triggering system was described in detail in McGillivray and Mayne [7]. Compared to conventional SCPTu testing that allows V s at standard 1-m depth intervals, non-stop V s measurement system of a https://doi.org/10.1016/j.measurement.2020.107610 0263-2241/Ó 2020 Elsevier Ltd. All rights reserved. ⇑ Corresponding author. E-mail addresses: [email protected](S.-W. Moon), [email protected]. kr (R.E. Kim), [email protected](A.C.C. Cheng), [email protected](Y.E. Li), [email protected](T. Ku). Measurement 156 (2020) 107610 Contents lists available at ScienceDirect Measurement journal homepage: www.elsevier.com/locate/measurement
Sung-Woo Moon a, Robin E. Kim b, Arthur C.C. Cheng c, Yunyue Elita Li c, Taeseo Ku c,⇑aDepartment of Civil and Environmental Engineering, Nazarbayev University, 53 Kabanbay Batyr Ave, Nur-Sultan 010000, KazakhstanbDepartment of Civil and Environmental Engineering, Hanyang University, Wangsimri-ro 222, Sungdong-gu, Seoul 15588, South KoreacDepartment of Civil and Environmental Engineering, National University of Singapore, 1 Engineering Drive 2, Singapore 117576, Singapore
a r t i c l e i n f o
Article history:Received 3 September 2019Received in revised form 12 January 2020Accepted 10 February 2020Available online 13 February 2020
Keywords:Seismic piezocone test (SCPTu)Background noiseCross power spectral densityCone tip resistanceSleeve friction
a b s t r a c t
The continuous-interval seismic piezocone test (CiSCPTu) system has been recently developed for obtain-ing detailed information on shear wave velocity (Vs) along the depth. In the seismic signal processing,background noises are generally unfavorable since it can mask main shear waves. However, this studyintroduces a novel idea to utilize the unwanted noises to characterize soil types. The method is basedon the integrated and normalized cross power spectral density (bCyk ) of the background noises, calculatedusing a chosen reference signal, and the cone measurements (i.e., cone tip resistance (qt) and sleeve fric-tion (f s)). We used the background noise recordings collected at Richmond (Canada) with well-definedgeological layers. The developed correlations between background noise and cone measurements are ver-ified via two traditional CPT-based Soil Behavior Type (SBT) classification systems. The results show thatthe simplified soil types into sand and clay in SBT system predicted by the developed relationships (i.e.,92 to 95% of sand and 60 to 88% of clay) are comparable with the identification (i.e., 96 to 97% of sand and69 to 99% of clay) with CPT measurements.
� 2020 Elsevier Ltd. All rights reserved.
1. Introduction
Since the in-situ shear wave velocity (V s) is a critical parameterfor determining the initial shear stiffness (G0 or Gmax), required fordealing with the geotechnical engineering problems such as staticand dynamic deformations and liquefaction potentials, variousgeophysical methods have been utilized to obtain the in-situ V s.For example, invasive techniques such as downhole test (DHT)and crosshole test (CHT), and non-invasive techniques such asSpectral Analysis of Surface Waves (SASW) method [1] andMulti-Channel Analysis of SurfaceWaves (MASW) method [2] havebeen predominantly employed in exploration geophysics as well asgeotechnical engineering applications.
The conventional DHT and CHT tests are much more costly andtime-consuming than the non-invasive techniques, albeit they pro-vide reliable in-situ V s. Alternatively, direct push downhole-typemethods such as seismic cone penetration test (SCPT) and seismicflat dilatometer test (SDMT) can produce equivalent promisingresults for determining in-situ V s profiles [3–5]. Especially, the
SCPT may be the most cost-efficient method and became an attrac-tive option when comparing typical costs of several geophysicaltests [6]. Fig. 1(a) and (b) present the general setup and proceduresof SCPT or SDMT to measure in-situ V s, including two types of com-mon approaches such as a pseudo-interval method using a singlereceiver (Fig. 1(a)), and a true interval method using multiplereceivers (Fig. 1(b)). While a seismic peizocone is pushed into theground, in-situ V s generated from a horizontal seismic source ismeasured at a single or dual seismic receiver generally per every1 m-depth interval. In addition, cone tip resistance (qt), sleeve fric-tion (f s), and pore-water pressure (u2) with depth are providedtogether from a single sounding.
Based on the concept of an automatic mechanical gear system, anew portable automated seismic triggering source (i.e., RotoAuto-Seis), developed by McGillivray and Mayne [7], allows measuringthe continuous variation of shear wave velocity (V s) with depth,as well as typical readings of piezocone penetration test (CPTu)(i.e., qt , f s, u2) during non-stopping cone penetration. Fig. 1(c) illus-trates the automated impulse source that can generate continuousshear waves consistently and repeatedly at a constant rate of speed(e.g., every 1 to 10 s). The schematic diagram of the triggeringsystem was described in detail in McGillivray and Mayne [7].Compared to conventional SCPTu testing that allows Vs at standard1-m depth intervals, non-stop Vs measurement system of a
CiSCPTu Continuous-interval seismic piezocone testVs Shear wave velocityqt Cone tip resistancefs Sleeve frictionSBT Soil behavior typeG0 or Gmax Initial shear stiffnessDHT Downhole testCHT Crosshole testSASW Spectral analysis of surface wavesMASW Multi-channel analysis of surface wavesSCPT Seismic cone penetration testSDMT Seismic flat dilatometer testRotoAutoSeis Automated seismic triggering sourcePSD Power spectral density
Rykyl Cross-correlation functionCPSD Cross power spectral densitybCyk Integrated and normalized cross power spectral densitybqt Normalized cone tip resistancebf s Normalized sleeve frictionCRl Minimum squared residual between bCyk and bqt
SRl Minimum squared residual between bCyk and bf sR2 A coefficient of determinationS.E.Y. A residual standard error in regressionQtn Normalized cone tip resistanceFr Sleeve friction ratio
2 S.-W. Moon et al. /Measurement 156 (2020) 107610
continuous-interval SCPTu (CiSCPTu) can result in the significanttime reduction of field testing (minimal stopping required onlyat rod break) and enhanced resolution of the site characterization(detailed Vs profile). The most challenging part of CiSCPTu testingis not only recording the vibration of cone caused by the penetra-tion, but also analyzing/interpreting the acquired seismic data (i.e.,signal and background noise) due to the sensitivity influenced byseveral factors such as noisy signals, reflected and refracted shearwaves, and extraordinarily short travel times. To obtain reliablein-situ Vs measurements from CiSCPTu, appropriate signal post-processing methods have been discussed in both the time domainand the frequency domain [6,8,9]. For example, acquired continu-ous Vs data from a test site in southeast Virginia were interpretedby using cross-correlation in the time domain and spectral densityanalysis in the frequency domain with a zero-phase distortion fil-tering technique to mitigate the sensitivity of Vs evaluations [6].
Generally, noise is considered as a redundant signal superim-posed on the desirable signal, thus several signal filtering methods(e.g., stacking, moving average with a smoothing kernel, low/high-pass filter and band-pass filter) have been employed for reducingthe noise level. On the contrary, seismic noise recordings from nat-ural and cultural noises have been used for developing horizontal-to-vertical (H/V) spectral ratio curves and dispersion curves of sur-face waves [10–15], and estimating a suitable characterization oflocal site effects of the sedimentary covers [16]. In this study, anew and practical link between background noises measured fromCiSCPTu and cone measurements (i.e., qt and f s) is investigated for
R1
R2
Z1Z2 t1
t2
Seismic source: triggering at given depth
R12 = z12 + x2
R22 = z22 + x2
x
Vs = ∆R / ∆t
Pseudo-interval seismic system
R1
R2
t1
t2
x
True-interval seismic system
Single seismic receiver
Alterna�ng sequence: CPT + DHT
Dual seismic receiver
(a) (b)
Fig. 1. Three different direct-push seismic cone penetration systems for shear wave vesystem, and (c) continuous-interval seismic system.
soil type classification. It is expected that the proposed methodol-ogy can add considerable values to unwanted background noisescaptured during cone penetration.
2. Site Description: Richmond, British Columbia
The test site for CiSCPTu is located at an industrial site in Rich-mond, British Columbia, Canada. From a geological map, it is foundthat subsurface of the test site in Richmond is underlain by Quater-nary deposits and formed by a typical topset sequence, in whichthe sand layer is 8 to 25 m thick [17]. Below the topset sand layer,tidal flat and bar top environments lead to interbedded sand siltdeposited, which is overlain by clayey organic silts deposited inthe tidal marsh and floodplain environments [18,19]. Based onthe site investigation, the ground condition at the test site in Rich-mond is mainly composed of four different geomaterial types: (1)gravelly sand fill at depths of 0 to 1 m; (2) natural alluvial and del-taic deposits of silty clay at depths of 1 to 7 m; (3) a thick sandlayer at depths of 7 to 30 m; and (4) a thick layer of soft to firmclayey silt at depths of 30 to 45 m. Depth to groundwater table isabout 3.5 m at a test site location.
In order to conduct CiSCPTu, a 15 cm2 steel cone with a singlegeophone located 20 cm above the cone tip (i.e., a pseudo-interval seismic penetrometer) was penetrated into the ground.Fig. 2 presents the measured typical piezocone readings such astip resistance (qt), sleeve friction (f s), and pore pressure (u2) at
Fig. 2. Piezocone soundings at Richmond, BC site: (a) cone tip resistance, (b) sleevefriction, (c) excess pore water pressure [20].
S.-W. Moon et al. /Measurement 156 (2020) 107610 3
the test site. Moreover, an automatic seismic source (namedRotoAutoSeis), located on the ground with a horizontal offset of1.25 m from the CPT rod string, generated continuous shear wavesevery 5 s during penetration of the steel cone. In total, 445 shearwave signals with an entire record time of 0.4 s were successfullygenerated and measured at each 10 cm vertical interval up to 45 m.Fig. 3 shows the continuous record of raw wavelets with clearidentification of shear waves (Fig. 3(a)) and the continuous Vs pro-file evaluated from cross-spectral analysis in frequency domain(Fig. 3(b)). All seismic raw data were obtained from ConeTechInvestigations, Inc. In Fig. 3(b), a reference true-interval DHT Vs
profile was also presented with the continuous Vs profile for com-parison. A detailed procedure for estimating the continuous Vs pro-file at this site was described in Ku and Mayne [20].
3. Proposed signal processing methodology
3.1. Characterization of background noise
Commonly, random signals can be classified into two types,such as stationary and non-stationary signals, depending on the
0
5
10
15
20
25
30
35
40
45
500 50 100 150 200 250
Time (ms)
Dep
th (m
)
(a)
Fig. 3. (a) Continuous raw shear wave signals recorded from CiSCPTu conducted in Richmraw shear wave signals.
changes of statistical properties with the time [21]. To identifythe characteristics of background noise components captured inthe shear wave signals, we attempted to isolate the signals bytrimming (note that a length of trimmed signal is 200 ms) asshown in Fig. 4. Then, statistical parameters (i.e., mean and vari-ance) were determined in the remained signals, which obviouslyexclude the main shear wave signals of high amplitudes [21].Fig. 4 presents the distributions of the signal amplitudes, whichis computed using the ensemble of the signals, corresponding toarbitrary times (e.g., t = 50 ms, 150 ms, 250 ms, and 350 ms)and depths (e.g., d = 8 m, 24 m, and 40 m). The results show thatthe measured shear wave signals without high amplitudes areapproximately normally distributed with zero means and vari-ances at each time and depth, respectively. Thus, it may implythat the background noise parts of the measured shear wave sig-nals are stationary and ergodic [21].
Interestingly, however, it was found that the level of observedbackground noises (e.g., signal amplitude and fluctuation) is actu-ally not constant within the profile. Our approximate visual obser-vation indicates that the background noise level seems to varydepending on soil types at certain depths. For example, therecorded background noise amplitudes in 10 to 30 m depths (sandlayer) were much larger than those observed in 30 to 45 m depths(clay layer). Probably the background noises are attributed tovibration of cone rod as it penetrates down the subsurface soil.Sand, being generally rougher and stiffer than clay, is expected toprovide more penetration resistance and cause stronger vibration.This can justify a possible link between the background noise char-acteristics and penetration or frictional resistances in different par-ticular materials.
Based on the insight, the measured seismic data are post-processed to investigate how the background noise componentsare extracted and used for estimating the soil type in a systematicmanner. Fig. 5(a) shows the raw shear wave signals in threedimension. The raw signals were collected at 20 kHz with a recordlength of 400 ms and contained various signal components; theresponse of the soil due to input seismic wave, input and outputbackground noise, etc. Before starting signal post-processing,input-dependent signals striking rich in the shear modes must beeliminated to utilize the measured background noises for charac-terizing the subsurface soils.
One option for eliminating the strike involving shear compo-nents, which may be correlated with the input signal, is to passthe signal through an appropriately designed filter. The peaks from
300 350 400
0
5
10
15
20
25
30
35
40
45
50
0 100 200 300 400
Dep
th (m
)
Vs (m/sec)
True Interval Vs(Reference)
(b)
ond, BC, and (b) continuous shear wave velocity profile developed from continuous
0
5
10
15
20
25
30
35
40
45
50
Dep
th (m
)
0 50 100 150 200 250 300 350 400Time (ms)
-1-0.5
00.5
10
100
200
300
400
500
600
Am
plitude
Frequency
Distribution at d=40m
-0.50
0.50
100
200
300
400
500
600
Am
plitude
Frequency
Distribution at d=24m
-0.50
0.50
100
200
300
400
500
600
Am
plitude
Frequency
Distribution at d=8m
-1 -0.5 0 0.5 10
20
40
60
80
100
Amplitude
Freq
uenc
y
Distribution at t=50ms
-1 -0.5 0 0.5 10
20
40
60
80
100
120
AmplitudeFr
eque
ncy
Distribution at t=150ms
-1 -0.5 0 0.5 10
20
40
60
80
100
Amplitude
Freq
uenc
y
Distribution at t=250ms
-1 0 10
50
100
150
Amplitude
Freq
uenc
y
Distribution at t=350ms
Fig. 4. Normal distribution of the ensemble of raw signals that exclude the range of 200 ms including high amplitudes.
4 S.-W. Moon et al. /Measurement 156 (2020) 107610
the shear strikes contain rather low-frequency signals, and thuscan be eliminated via passing a high-pass filter. Ku et al. [6]reported that shear waves triggered by the auto-source show thehighest coherence at a frequency range of about 50 and 100 Hz.Herein, a cut-off frequency of the high-pass filter is set at 150 Hzto eliminate the shear waves.
Fig. 5(b) shows the auto power spectral density (PSD) at eachdepth. Along the entire depth of the soil, higher peaks are observedin the frequency domain of 3,000–4,000 Hz, and 7,000–8,000 Hz,indicating that either soil or input related components exist atthe range between. If those frequencies (at the higher frequencyarea) are related to the input, lower frequency area might beaffected (corrupted) by the aliasing. The upper end of the fre-quency range must be selected to minimize the error in estimatingthe soil characteristics. Thus, the upper end of the frequency hasbeen set at 1,000 Hz by resampling the data.
Thereafter, to better visualize Fig. 5(a) and (b), time signals areaveraged in Fig. 5(c). Although the amplitude is reduced andsmoothed, most prevalent signal components are remained. Then,using stacked (i.e., averaged) signal, the auto PSD is plotted in Fig. 5(d). In the figure, the higher energy is observed in the lower fre-quency range while the first plateau is found around 150 Hz. Thus,the frequency range beyond 150 Hz is considered as the back-ground noise signal in the rest of the paper. It also verifies thatthe selected cut-off frequency (i.e., 150 Hz) of high-pass filter isappropriate.
Fig. 5(e) and (f) show the post-processed raw shear wave sig-nals in time domain and the frequency domain. In calculating theauto PSD, the number of Fast Fourier Transform (FFT) points is1,024, and hanning window is used. As can be seen, shear strikes
are removed by the high-pass filter, and depth-dependent autoPSD are observed at the frequency range of 0–500 Hz.
3.2. Cross power spectral density (CPSD)
Under random vibration, soil properties have been estimated byseveral signal processing techniques such as the power spectraldensity function, the autocorrelation function, and the transferfunction [22–24]. This study examines the possibility of using crosspower spectral density (CPSD) to identify the soil types along thedepth using background noises from CiSCPTu, as a new trialtowards a quantitative approach. A hypothesis is that correlatedcharacteristics between background noises may vary dependingon different types of soil layers due to different features of degreeof cone vibration or generated energy. In addition, the effect ofunwanted signals (e.g., remained white noise components, shearmode peaks) at all frequency ranges, which cannot be eliminatedby high-pass filter, can be efficiently reduced by using CPSD [25–27]. To consider such effect, the reference signal (i.e., the denomi-nator signal) needs to be chosen carefully [25]. If the measured andprocessed signals from depth at k and l are yk tð Þ and yl tð Þ, respec-tively, and the equations can be defined as follows:
yk tð Þ ¼ xk tð Þ þ nxk tð Þ ð1Þ
yl tð Þ ¼ xl tð Þ þ nxl tð Þ ð2Þ
where nxk tð Þ and nxl tð Þ are the measured noise components of vari-ous sources. Also, xk tð Þ and xl tð Þ are signals containing (1) the back-ground noise level related to soil characteristics and (2) the
Nor
mal
ized
ra
w si
gnal
Time (sec)Frequency (Hz)
Raw
sign
al
PSD
(dB
)
(a) (b)
(c) (d)
Time (sec)
Res
ampl
ed
sign
al Res
ampl
ed
sign
al
PSD
(dB
)
Frequency (Hz)(e) (f)
0 50 100 150 200 250 300 350 400 450 500-80
-70
-60
-50
-40
Stac
ked
raw
si
gnal
PSD
(dB
)
Frequency (Hz)
Stac
ked
raw
sign
al
0 50 100 150 200 250 300 350 400-0.05
0
0.05
0.1
0.15
Time (sec)
Fig. 5. Measured signal from CiSCPTu (a) raw signal in time domain; (b) auto PSD of the raw signal; (c) stacked raw signal in time domain; (d) auto PSD of the stacked rawsignal; (e) processed signal in time domain; (f) auto PSD of the processed signal. (Note: The x-axis is the depth in meters, y-axis is time (milliseconds), and the z-axis is theraw signal normalized by a maximum magnitude of each signal).
S.-W. Moon et al. /Measurement 156 (2020) 107610 5
remaining shear wave after filtering. Then, the cross-correlationfunction (Rykyl ) between the both signals (yk tð Þ and yl tð Þ) can be esti-mated as:
Rykyl sð Þ ¼ E yk t þ sð Þyl tð Þ½ �¼ Rxkxl sð Þ þ Rxknxl
sð Þ þ Rxlnxksð Þ þ Rnxk nxl
sð Þ ð3Þ
where s is the time lag.The correlation function in equation (3) will be the most accu-
rately estimated when an ideal reference signal that contains thesmallest signal noise (i.e., nxl tð Þ) or represents soil characteristicsis chosen. To determine the reference signal, covariance (C) ofthe processed signal is defined by integrating each CPSD(Sykyl xð Þ) in frequency domain transferred from correlation func-tion (Rykyl sð Þ) in time domain over the frequency region of theinterest as follows:
C ¼Z 500= 2pð Þ
0Sykyl xð Þdx ð4Þ
where l ¼ 1; � � � ;N, N is the total number of measurement along thesoil profile, x is the angular frequency. In this specific case, sincethe high-pass filter was designed to contain only signals over150 Hz, the integration was also made up to 500 Hz.
It is well-known that generally the magnitude of cone tip resis-tance (qt) is strongly correlated to particle characteristics or soiltypes. Here, the minimum squared residual (CRl) between the esti-
mated covariance at yk-th column of C, (herein noted as Cyk ) andthe measured qt is calculated to assure that the chosen referencesignal contains the lowest background noise or well contains thesoil characteristics:
CRl ¼ minXNk¼1
bCyk � bqt
� �2 !ð5Þ
where qt is the cone tip resistance, and a hat indicates the normal-ized quantity. Because Cyk and qt have different units, Cyk is also nor-malized signals by the maximum among the signals and used forcalculating the squared residuals.
Fig. 6(a) shows the integrated and normalized CPSD (bCyk ) ofthe processed background noise, using equation (4), where thereference signal gives the smallest CRl in equation (5), along withthe normalized cone tip resistance (bqtÞ. Two black horizontallines indicate the borders of different soil types as presented inFig. 2; the upper and lower layers are predicted to be the clay,while the middle layer is predicted to be the sand layer. ThebCyk of the measured signal shows rapid increase at the sandlayer, so as in bqt . Here, the selected reference is the signal at20 m and denoted with a dot in Fig. 6(b). Although some phasesseem to be off around 15 m, 20 m, and 26 m, the applied back-ground noise-based approach using only the measured back-ground noise can predict the changes of soil type. Further, to
0.756
(a) (b)
Cone tip resistance (qt)
Silt & Clay
Sand
Silty clay
Silt & Clay
Sand
Silty clay
Fig. 6. (a) Comparison of integrated and normalized CPSD (bCyk ) of measured background noise and normalized cone tip resistance, and (b) correlation coefficient of integratedand normalized CPSD (bCyk ) of measured background noise and normalized cone tip resistance.
0.655
(a) (b)
Sleeve friction (fs)
Silt & Clay
Sand
Silty clay
Silt & Clay
Sand
Silty clay
Fig. 7. (a) Comparison of integrated and normalized CPSD (bCyk ) of measured background noise and normalized sleeve friction, and (b) correlation coefficient of integrated andnormalized CPSD (bCyk ) of measured background noise and normalized sleeve friction.
6 S.-W. Moon et al. /Measurement 156 (2020) 107610
demonstrate how a reference signal can be selected for thefuture applications, Fig. 6(b) plots an example of the correlation
coefficient of the bCyk of the measured background noise and thebqt when the reference signal is selected at a specific depth asfollows:
qk ¼ corrcoeficient bCyk ; bqt
� �ð6Þ
where k = 1,. . .,N. It was observed that the sand layer provides aboutthe correlation coefficient of 0.756, indicating the any middle leveldepth can produce similar degree of soil layer estimation.
0
20
40
60
80
100
0.01 0.1 1
Sle
eve
frict
ion,
f s(k
Pa)
Normalized and integrated CPSD (ĉyk)
fs (kPa) = 64.8(ĉyk)0.69
R2 = 0.754
Clay
SandMean ĉyk
between 7 m and 8 m
Mean ĉykbetween 30 m and 31 m
0
5,000
10,000
15,000
20,000
0.01 0.1 1
Con
e tip
resi
stan
ce, q
t(k
Pa)
Normalized and integrated CPSD (ĉyk)
qt (kPa) = 15,305(ĉyk)0.81
R2 = 0.733
SandMean ĉykbetween 7 m and 8 m
Mean ĉykbetween 30 m and 31 m
Clay
(a) (b)
Fig. 8. Relationships between integrated and normalized CPSD (bCyk ) and (a) cone tip resistance, or (b) sleeve friction.
0
10
20
30
40
50
60
70
0.01 0.1 1
Nor
mal
ized
con
e re
sist
ance
, Qtn
Normalized and integrated CPSD (ĉyk)
Qtn = 70.0(ĉyk)1.10
R2 = 0.871 Sand
Mean ĉykbetween 7 m and 8 m
Mean ĉykbetween 30 m and 31 m
Clay
0
0.5
1
1.5
2
0.01 0.1 1
Nor
mal
ized
fric
tion
ratio
, Fr
Normalized and integrated CPSD (ĉyk)
Fr = 0.4(ĉyk)-0.29
R2 = 0.906
Clay
Sand
Mean ĉykbetween 7 m and 8 m
Mean ĉykbetween 30 m and 31 m
(a) (b)
Fig. 9. Relationships between integrated and normalized CPSD (bCyk ) and (a) normalized cone tip resistance, or (b) normalized sleeve friction.
S.-W. Moon et al. /Measurement 156 (2020) 107610 7
Similarly, the approach can be also applied to sleeve frictionmeasurements (f s) by calculating the minimum residual square
(SRl) between the bCyk and the normalized sleeve friction (bf s), asfollows:
SRl ¼ minXNk¼1
bCyk � bf s� �2 !ð7Þ
where l = 1,. . .,N. The location of SRl was found at l = 20 m.
Fig. 7(a) presents the comparison of the estimated bCyk of themeasured background noise and the normalized sleeve friction
(bf s). Note that the normalization is used to make both parameters
dimensionless. The results also verify that the bCyk of the measuredbackground noise can be used for estimating soil types, showinggood agreements with the sleeve friction at the borders of the soil
type. Fig. 7(b) plots the correlation coefficient of the estimated bCyk
and the bf s. The correlation coefficient of 0.655 is calculated at thedepth of 20 m which is the same location as the cone tip resistance
(qt) (Fig. 6(b)). It appears that the approach utilizing the bCyk in esti-mating the soil type is not very sensitive to the selected referencesignal, and any middle layer can be used for the reference signal.
4. Correlation between background noise and conemeasurements
Apparent trends between integrated and normalized cross
power spectral density (bCyk ) and measurements of cone tip
resistance (qtÞ, Fig. 8(a), or sleeve friction (f sÞ, Fig. 8(b), are investi-gated. A representative mean value at every 1 m-depth interval inbetween 7 m and 45 m is used for regressions to consider differentsampling depth intervals as well as to avoid undesirable sensitiveerrors. The trends are appropriately fitted with the followingexpressions:
qt kPað Þ ¼ 15;305 bCyk
� �0:81;n ¼ 38;R2 ¼ 0:733; S:E:Y: ¼ 0:236 ð8Þ
f s kPað Þ ¼ 64:8 bCyk
� �0:69;n ¼ 38;R2 ¼ 0:754; S:E:Y: ¼ 0:192 ð9Þ
where n = a total number of data, R2 = a coefficient of determina-tion, S.E.Y. = a residual standard error in regression. In terms of
the bCyk , the regression analyses provide good correlation coeffi-cients for (a) qt and (b) f s as shown in equations (8) and (9), respec-
tively. Fig. 8 presents that qt , f s, and bCyk data are clustered indifferent portions, depending on two soil types (i.e., sand and clay).In Fig. 8, qt and f s for clay have mainly lower values (maximum:1,524 kPa, and 14.91 kPa) than those for sand (minimum:
5,693 kPa, and 15.76 kPa), respectively. In addition, bCyk for clay
has the lower range between 0.05 and 0.20, compared to bCyk forsand which has the greater range between 0.20 and 0.86. Exception-
ally, low values of bCyk , qt , and f s (i.e., 0.01, 2,505 kPa, and 10.61 kPa,respectively) for sand in silt & clay to sand transition zone between
7 m and 8 m, and high values of bCyk and f s (i.e., 0.38 and 26.36 kPa,respectively) for clay in sand to clay transition zone between 30 mand 31 are noted. It appears that the transition zone of soil types
1
10
100
1000
0.1 1 10
Qtn
estim
ated
fro
m d
evel
oped
cor
rela
tions
Fr estimated from developed correlations
7 to 30 m using Eq. (8 & 9)30 to 45 m using Eq. (8 & 9)7 to 30 m using Eq. (12 & 13)30 to 45 m using Eq. (12 & 13)
1. Sensitive, fine grained
9. Very stiff,fine graned soils
8. Very stiff sand to clayey
sand7. Gravelly sandto dense sand
(b) Robertson (2004)
1
10
100
1000
0.1 1 10
Qtn
estim
ated
fro
m c
one
mea
sure
men
ts
Fr estimated from cone measurements
7 to 30 m30 to 45 m
1. Sensitive, fine grained
9. Very stiff, fine grained soils
8. Very stiff sand to clayeysand
7. Gravelly sand to dense sand
(a) Robertson (2004)
1
10
100
1000
0.1 1 10
Qtn
estim
ated
fro
m c
one
mea
sure
men
ts
Fr estimated from cone measurements
7 to 30 m30 to 45 m
c=0.15
c=0.35
c=0.55
c=0.75
c=1.0
1.Sands & Gravels
Normally consolidated
Cone resistnace stress exponent
Loose
Medium Dense
Dense
Very Dense
Organic clay and unstable clayey silt
(c) Olsen&Mitchell (1995)
1
10
100
1000
0.1 1 10
Qtn
estim
ated
fro
m d
evel
oped
cor
rela
tions
Fr estimated from developed correlations
7 to 30 m using Eq. (8 & 9)30 to 45 m using Eq. (8 & 9)7 to 30 m using Eq. (12 & 13)30 to 45 m using Eq. (12 & 13)
c=0.55
c=0.15
c=0.35c=0.75
c=1.0
1. Sands & Gravels
Normally consolidated
Cone resistnace stress exponent
Loose
Medium Dense
Dense
Very Dense
Organic clay and unstable clayey silt
(d) Olsen&Mitchell (1995)
Fig. 10. Comparison of soil behavior type (SBT) charts derived from the correlations based on (a, c) measurements of cone tip resistance and sleeve friction, and (b, d)background noise.
8 S.-W. Moon et al. /Measurement 156 (2020) 107610
may lead to the unexpected variations of bCyk , qt , and f s probably dueto uncertainty related to entry and exit effect of mixed backgroundsignal measurements and cone measurements. Nonetheless, we can
observe clear tendencies for qt and f s to increase with bCyk about twodifferent soil types.
Similarly, apparent trends between bCyk and normalized cone tipresistance (QtnÞ, Fig. 9(a), or sleeve friction ratio (FrÞ, Fig. 9(b), areexamined. Two stress-normalized CPT parameters (Qtn and Fr)are defined as follows [28]:
rv0 = total vertical overburden stress, r0v0 = effective vertical
stress.
Fig. 9 shows that Qtn, Fr , and bCyk data are clustered in differentportions, depending on two soil types (i.e., sand and clay).
Interestingly, using the normalized CPT parameters, even stronger
trends between bCyk and Qtn or Fr are observed regarding twodifferent soil types, compared to the apparent trends using qt
and f s. The trends are appropriately fitted with the followingexpressions:
Qtn ¼ 70:0 bCyk
� �1:10;n ¼ 38;R2 ¼ 0:871; S:E:Y: ¼ 0:206 ð12Þ
Fr ¼ 0:4 bCyk
� ��0:29;n ¼ 38;R2 ¼ 0:906; S:E:Y: ¼ 0:045 ð13Þ
5. Soil type classification – Comparison using SBT charts
For soil type identification, the possibilities of using the devel-oped relationships between cone tip resistance (qtÞ or sleeve fric-tion (f s) and integrated and normalized cross power spectral
density (bCyk ) are verified via Soil Behavioral Type (SBT) classifica-tion charts developed by Olsen and Mitchell [29] and Robertson[30], using two stress-normalized CPT parameters (Qtn and Fr).
0
50
100
150
200
250
Clay (1&3) Silt (4) Sand (5&6)
Cou
nt o
f dat
a po
ints
Soil behavior type in SBT chart
Measured qt & fsEq. 8 & 9Eq. 12 & 13
(a) Robertson (2004)
0
50
100
150
Clay (1&3) Silt (4) Sand (5&6)
Cou
nt o
f dat
a po
ints
Soil behavior type in SBT chart
Measured qt & fsEq. 8 & 9Eq. 12 & 13
(b) Robertson (2004)
0
50
100
150
200
250
Clay (5) Silt (4) Sand (2&3)
Cou
nt o
f dat
a po
ints
Soil behavior type in SBT chart
Measured qt & fsEq. 8 & 9Eq. 12 & 13
(c) Olsen&Mitchell (1995)
0
50
100
150
Clay (5) Silt (4) Sand (2&3)C
ount
of d
ata
poin
ts
Soil behavior type in SBT chart
Measured qt & fsEq. 8 & 9Eq. 12 & 13
(d) Olsen&Mitchell (1995)
Fig. 11. Comparison of the soil types predicted by using the proposed equations, and estimated by using qt and f s measurements at different depths: (a, c) 7 – 30 m and (b, d)30 – 45 m.
S.-W. Moon et al. /Measurement 156 (2020) 107610 9
Fig. 10 plots data in pairs of Qtn and Fr determined from thedeveloped equations (8 and 9), and equations (12 and 13), anddirectly from the measured qt and f s. Generally, the data inFig. 10(a) and (c) are clustered with scattering, compared withthe data forming lines in Fig. 10(b) and (d). It is attributed to thesource differences/characteristics because the original Qtn and Fr
are obtained based on independent measurements (i.e., qt andf s), while the newly developed correlations to calculate pairs of
Qtn and Fr are linked via bCyk . Using the SBT chart from Robertson[30], the data obtained using both the measured qt and f s, Fig. 10(a), and the developed equations, Fig. 10(b), mainly cover zone 5and 6 (sand mixtures and sands) in the depth of 7 m to 30 m,and zone 1 (sensitive, fine grained) in the depth of 30 m to 45 m.Employing the SBT chart from Olsen and Mitchell [29], the dataestimated from both the measured qt and f s, Fig. 10(c), and thedeveloped equations, Fig. 10(d), mainly cover zone 2 (sand) inthe depth of 7 m to 30 m, and zone 5 (clay) in the depth of 30 mto 45 m. It is demonstrated that the data estimated from back-ground noises cover similar portions of the SBT classification chartswith pairs of Qtn and Fr that determined from the direct CPTmeasurements.
Fig. 11 presents quantitative comparisons of the soil behaviortypes predicted by classic CPT-based identification using conemeasurements (qt and f s), and estimated by equations (8 and 9),and equations (12 and 13), respectively. In order to apparentlycompare the estimated soil types according to Robertson [30], ninesoil types of SBT system are simplified into three basic soil types:(1) clay including soil type zones of 1 and 3; (2) silt correspondingto a soil type zone of 4; (3) sand including soil type zones of 5 and6. The two soil behavior types (i.e., 97% sand in the depth of 7 m �30 m and 99% clay in the depth of 30 m � 45 m) are mainly pre-dicted by classic CPT-based identification using measured qt andf s. In the depth of 7 m to 30 m, a dominant soil type estimated fromthe proposed approach is equivalent as follows: 92% sand from
equations (8 and 9), and 93% sand from equations (12 and 13)(Fig. 11(a)). In the depth of 30 m to 45 m, the estimated soil typeis 84% clay from equations (8 and 9), and 88% clay from equations(12 and 13) (Fig. 11(b)).
For apparent comparison of the estimated soil types accordingto Olsen and Mitchell [29], three basic soil types are simplifiedwith: (1) sand including soil type zones of 2 and 3; (2) silt corre-sponding to a soil type zone of 4; (3) clay corresponding to a soiltype zone of 5. The two soil behavior types (i.e., 96% sand in thedepth of 7 m � 30 m and 69% clay in the depth of 30 m � 45 m)are mainly predicted by the classic CPT-based identification usingmeasured qt and f s. In the depth of 7 m to 30 m, a dominant soiltype estimated from the proposed approach is equivalent as fol-lows: 94% sand from equations (8 and 9), and 95% sand from equa-tions (12 and 13) (Fig. 11(c)). In the depth of 30 m to 45 m, theestimated soil type is 84% clay from equations (8 and 9), and 60%clay from equations (12 and 13) (Fig. 11(d)).
In summary, it is demonstrated that the basic soil types withpairs of Qtn and Fr identified from the proposed equations (pairsof 8 and 9, and 12 and 13) have good agreements with the classicCPT-based references that indicate sand types in the range of 7 mto 30 m, clay types in the range of 30 m to 45 m.
6. Discussion
In this study, using background noise used to be ignored or con-sidered as a redundant signal, the detailed procedure for character-izing soil type was introduced and examined based on (1)characterizing background noise; and (2) integrating and normal-
izing cross power spectral density (bCyk ) of the processed back-ground noise. In post-processing of auto source signal fromCiSCPTu, a proper filter among various types of filters (e.g., low-,high-, bandpass-filters) to eliminate or keep in selecting frequen-cies should be carefully chosen considering the characteristics of
10 S.-W. Moon et al. /Measurement 156 (2020) 107610
target signals. In addition, for the selection of the reference signalat the test site, it is noted that measured cone tip resistance (qt) orsleeve friction (f s), as can be seen in Fig. 2, exhibits a considerablecharacteristic variation with depth. It may lead to a significantinfluence on the results (e.g., generation of redundant results). Atthe testing site in this study, it was demonstrated that the proce-dure based on integrated and normalized cross power spectral
density (bCyk ) for estimating the soil type is insensitive to theselected reference signal, and the reference signal in any middlelayer could be used. Further investigation is strongly recom-mended in the future.
Typically, it is well recognized that the results obtained fromgeophysical testing data often contain considerable uncertaintyin interpreting site-specific conditions. Therefore, in practice, itshould be noted that the proposed approach can be implementedwith background noise obtained from continuous-interval seismicpiezocone test (CiSCPTu) in a test site with favorable soil character-istics which consist of well-defined geological layers. The proposedapproach may also require an appropriate background on relevantsignal processing techniques for implementation. Despite thenoted limitations, this study proposed the very interesting concep-tual idea (a proof of concept) to render undesirable backgroundnoise signals to be useful information and successfully demon-strated the feasibility with a case study.
7. Conclusions
Generally, the soil types have been commonly classified bymeans of various geotechnical and geophysical testing methods.In this study, we proposed a new background noise-basedapproach to characterize different soil types (i.e., sand and clay)using data from continuous-interval seismic piezocone test(CiSCPTu). We found three main conclusions as follows: (1) duringcontinuous cone penetration, background signal level measured ina geophone of a cone rod appears to vary with soil types; (2) when
integrated and normalized CPSD (bCyk ) of the measured backgroundnoise is plotted with the CPT measurements such as cone tip resis-tance (qtÞ, sleeve friction (f s), normalized con tip resistance (Qtn),and normalized sleeve friction (Fr), data are apparently separateddepending on the soil types (i.e., lower ranges for clay and greater
ranges for sand); (3) the bCyk based approach can provide the rea-sonable estimation of soil types (i.e., 92 to 95% of sand and 60 to88% of clay) from Soil Behavior Type (SBT) classification chartswhich are comparable with the identification (i.e., 96 to 97% ofsand and 69 to 99% of clay) from CPT measurements.
Based on the CiSCPTu data obtained from an ideal test site(showing clear separation between sand and clay layer) at Rich-mond, BC, we demonstrated that our newly developed backgroundnoise-based approach has the ability to be employed for estimatingsoil types directly using only background noise that we have com-monly considered undesirable. The developed background noise-based approach can be applied and considered reliable when thereare no chemical decomposition and structural changes of soil lay-ers. It may not be applicable for soil classification whereby the geo-logical conditions are rather unusual and complex. Nevertheless,the proposed approach motivates us to shift our thinking aboutbackground noise obtained from CiSCPTu, which has been consid-ered as unnecessary data. In the long-term study, because the testresults obtained from CiSCPTu may be site-specific, it should beapplied to multiple test sites with a wider range of soil types forcalibration and validation to reach more generalized results. Inaddition, a more sophisticated soil classification system could bedeveloped by building an extensive database including variousgeomaterials and geologic formations.
Authors would like to acknowledge that seismic cone penetra-tion test data were obtained from ConeTec Investigations, Inc.
References
[1] S. Nazarian, K. Stokoe, In situ shear wave velocities from spectral analysis ofsurface waves, Prentice Hall, San Francisco, California, July, 1984, pp. 21–28.
[2] C.B. Park, R.D. Miller, J. Xia, Multichannel analysis of surface waves, Geophysics64 (1999) 800–808.
[3] P.K. Robertson, R. Campanella, D. Gillespie, A. Rice, Seismic CPT to measurein situ shear wave velocity, J. Geotechnical Eng. 112 (1986) 791–803.
[4] G. Martin, P. Mayne, Seismic flat dilatometer in Piedmont residual soils, Proc.1st Int Conf. on Site Characterization, Atlanta (1998) 837–843.
[5] S. Marchetti, P. Monaco, G. Totani, D. Marchetti, In situ tests by seismicdilatometer (SDMT), From research to practice in geotechnical engineering 180(2008) 292–311.
[6] T. Ku, P.W. Mayne, E. Cargill, Continuous-interval shear wave velocity profilingby auto-source and seismic piezocone tests, Can. Geotech. J. 50 (2013) 382–390.
[7] A. McGillivray, P. Mayne, An automated seismic source for continuouspushshear wave velocity profiling with SCPT and frequent-interval SDMT, in:Proceedings of ISC-3: Geotechnical & Geophysical Site Characterization, 2,Taylor & Francis Group, TaipeiLondon, 2008, pp. 1347–1352.
[8] T. Ku, I. Weemees, E. Cargill, P.W. Mayne, D. Woeller, Post-processingcontinuous shear wave signals taken during cone penetrometer testing,Geotech. Test. J. 36 (2013) 543–552.
[9] M.A. Styler, J.A. Howie, J.T. Sharp, Perpetual source SCPTu: signal stacking shearwaves during continuous penetration, in: CPT’14 3rd International Symposiumon Cone Penetration Testing, Las Vegas, Nevada, USA, 2014.
[10] D. Fäh, F. Kind, D. Giardini, A theoretical investigation of average H/V ratios,Geophys. J. Int. 145 (2001) 535–549.
[11] H. Arai, K. Tokimatsu, S-wave velocity profiling by inversion of microtremor H/V spectrum, Bull. Seismol. Soc. Am. 94 (2004) 53–63.
[12] S.-W. Moon, Q. Khan, T. Ku, Application of MASWMethods for Investigations ofShear Wave Velocity in Residual Soils of Singapore, Geotechnical andStructural Engineering Congress 2016 (2016) 1688–1699.
[13] S.-W. Moon, K. Hayashi, T. Ku, Estimating Spatial Variations in Bedrock Depthand Weathering Degree in Decomposed Granite from Surface Waves, J.Geotechincal Geoenviron. Eng. 143 (2017) 04017020.
[14] S.-W. Moon, P. Subramaniam, Y. Zhang, G. Vinoth, T. Ku, Bedrock depthevaluation using microtremor measurement: empirical guidelines atweathered granite formation in Singapore, J. Appl. Geophys. 171 (2019)103866.
[15] S.-W. Moon, T. Ku, Estimation of bedrock locations and weathering degreeusing shear wave velocity-based approach, in: Proceedings of the 19thInternational Conference on Soil Mechanics and Geotechnical Engineering(ICSMG), Seoul, South Korea, 2017.
[16] M. Picozzi, A. Strollo, S. Parolai, E. Durukal, O. Özel, S. Karabulut, J. Zschau, M.Erdik, Site characterization by seismic noise in Istanbul, Turkey, Soil Dyn.Earthquake Eng. 29 (2009) 469–482.
[17] P.A. Monahan, V.W. Levson, B. Kerr, Geoscience Map 2010–2: QuaternaryGeology of Richmond, British Columbia (parts of 092G/02 and 03), BCGeological Survey, Victoria, BC, 2010.
[18] P. Monahan, P. Byrne, B. Watts, Naesgaard E., Engineering geology and naturalhazards of the Fraser River Delta. Guidebook for geological field trips inSouthwestern British Columbia and Northern Washington, Proceedings of theannual meeting of the Cordilleran section of the geological society of America,Vancouver, 2000, pp. 27-47.
[19] H.F. Williams, M.C. Roberts, Holocene sea-level change and delta growth:Fraser River delta, British Columbia, Can. J. Earth Sci. 26 (1989) 1657–1666.
[20] T. Ku, P. Mayne, Frequent-interval SDMT and continuous SCPTu for detailedshear wave velocity profiling in soils, Geotechnical Eng. J. SEAGS AGSSEA 43(2012) 34–40.
[21] J.C. Santamarina, D. Fratta, Discrete signals and inverse problems: anintroduction for engineers and scientists, John Wiley & Sons, 2005.
[22] F. Amini, K. Tawfiq, M. Aggour, Cohesionless soil behavior under randomexcitation conditions, J. Geotechnical Eng. 114 (1988) 896–914.
[23] M.S. Aggour, M.R. Taha, K.S. Tawfiq, F. Amini, Cohesive soil behavior underrandom excitation conditions, Geotech. Test. J. 12 (1989) 135–142.
S.-W. Moon et al. /Measurement 156 (2020) 107610 11
[24] F. Amini, K. Tawfiq, M. Aggour, Damping of sandy soils using autocorrelationfunction, in: Proceedings of the 8th Symposium on Earthquake Engineering, I,Roorkee, India, 1986, pp. 181-188.
[25] T. Nagayama, B.F. Spencer Jr, Structural health monitoring using smart sensors,Newmark Structural Engineering Laboratory, University of Illinois at Urbana,2007.
[26] H. Jo, S.-H. Sim, T. Nagayama, B. Spencer Jr, Development and application ofhigh-sensitivity wireless smart sensors for decentralized stochastic modalidentification, J. Eng. Mech. 138 (2011) 683–694.
[27] R.E. Kim, F. Moreu, B.F. Spencer, System identification of an in-service railroadbridge using wireless smart sensors, Smart Struct. Syst. 15 (2015) 683–698.
[28] P. Robertson, Interpretation of cone penetration tests-a unified approach, Can.Geotech. J. 46 (2009) 1337–1355.
[29] R.S. Olsen, J.K. Mitchell, CPT stress normalization and prediction of soilclassification, International Symposium on Cone Penetration Testing (CPT’95),Linkoping, Sweden, 1995, pp. 251–256.
[30] P. Robertson, Evaluating soil liquefaction and post-earthquake deformationsusing the CPT, Proc. Conf. on Site Characterization ISC (2004) 233–249.
