1. Introduction
Flow around bluff bodies has been extensively investigated over the past few decades due to its broad range of engineering applications such as in wind turbine towers, tubular heat exchangers and marine risers. Following the discovery of the Kármán vortex street, the topic of flow around cylindrical bodies, as a typical bluff body flow, has become an area of intensive research. The focus has been primarily on the flow dynamic characterisation, originating from the alternative vortex shedding, for circular cylinders, with numerous experimental (Cantwell & Coles Reference Cantwell and Coles1983; Triantafyllou, Triantafyllou & Chryssostomidis Reference Triantafyllou, Triantafyllou and Chryssostomidis1986; Kourta et al. Reference Kourta, Boisson, Chassaing and Minh1987; Zhou & Antonia Reference Zhou and Antonia1993; Norberg Reference Norberg1994; Casalino & Jacob Reference Casalino and Jacob2003) and numerical (Henderson Reference Henderson1997; Inoue & Hatakeyama Reference Inoue and Hatakeyama2002; Catalano et al. Reference Catalano, Wang, Iaccarino and Moin2003; Dong & Karniadakis Reference Dong and Karniadakis2005) studies continuously conducted over many years.
The family of polygonal cylinders, having quasi-axisymmetric cross-sectional geometries, share similar flow dynamic characteristics with the circular cylinder, in terms of Kármán type of vortex shedding when the flow regime is above the first critical Reynolds number. However, the side number
$N$
of the polygonal cylinder introduces distinct features which strongly influence the flow behaviour. Specifically, the periodic arrangement of flat surfaces and edges in the polygonal cylinder significantly affects the boundary layer development, the location of the separation points and the shear-layer separation angles (Xu et al. Reference Xu, Zhang, Gan, Li and Zhou2017). Geometric changes caused by
$N$
(Wang et al. Reference Wang, Xu, Gan, Zhang and Zhou2019, Reference Wang, Gan, Xu and Zhou2020), combined with the influence of incidence angle due to the non-axisymmetric shape (Masoudi, Gan & Sims-Williams Reference Masoudi, Gan and Sims-Williams2021; Cheng et al. Reference Cheng, Masoudi, Liu and Gan2024) lead to a highly nonlinear variation of vortex-shedding behaviours. Even though the shear layers tend to separate at fixed locations along two edges on the polygonal cylinder, one on each side in the transverse direction, and the Reynolds number
${\textit{Re}}$
is insensitive in the subcritical regime, they exhibit a stronger oscillatory flapping motion compared with that of the circular cylinder in a similar
${\textit{Re}}$
flow, giving rise to a more dynamic separation–reattachment mechanism (Masoudi, Sims-Williams & Gan Reference Masoudi, Sims-Williams and Gan2023). This complexity makes predicting the wake flow and the fluid forces acting on the polygonal cylinder particularly challenging.
From the flow control perspective, the polygonal cylinder introduces periodic geometric modification to the circular cylinder, and therefore can be considered as a passive control strategy (Zdravkovich Reference Zdravkovich1997; Choi, Jeon & Kim Reference Choi, Jeon and Kim2008). Experiments showed that polygonal cylinders with
$N\lt 8$
, particularly those with free ends, may undergo spontaneous rotation, generating a greater instantaneous lift force than the circular cylinder under a similar flow condition (Skews Reference Skews1991, Reference Skews1998). Bosch & Guterres (Reference Bosch and Guterres2001) investigated a tapered octagonal cylinder (
$N=8$
) and found that the drag coefficient in the face orientation, where a surface is facing the incoming flow normally, is lower than that of the corner orientation, where an edge faces the incoming flow, with the degree of tapering playing a critical role in aerodynamic performance. Tian, Niu & Wu (Reference Tian, Niu and Wu2011) studied a polygonal cylinder of
$N=24$
and reported a 40 % reduction in drag, accompanied by lower fluctuation levels, compared with the circular cylinder under similar flow conditions. A systematic experimental study by Xu et al. (Reference Xu, Zhang, Gan, Li and Zhou2017) for polygonal cylinders of
$N\in {\mathbb{Z}}[2,16]$
in face and corner orientations showed that, while polygonal cylinders with
$N\leqslant 8$
do not tend to undergo a transition from laminar to turbulent flow up to
$\textit{Re}=6 \times 10^{4}$
, for
$N\geqslant 12$
the discrete
$N$
corners can be asymptotically understood as
$N$
roughness elements evenly distributed around the circular cylinder surface with the characteristic roughness height inversely proportional to
$N$
, subsequently decreasing the critical
${\textit{Re}}$
for the laminar-to-turbulent transition.
According to Curle’s acoustic analogy (Curle Reference Curle1955; Phillips Reference Phillips1956), the periodic nature of vortex shedding from a cylinder into the wake region induces pressure fluctuations on the cylinder surface and hence produces acoustic waves, well known as the Aeolian tone. While a few studies have been conducted on the aerodynamic noise from square cylinders which are commonly seen in practice (Moreau & Doolan Reference Moreau and Doolan2013; Porteous, Moreau & Doolan Reference Porteous, Moreau and Doolan2017), there has not been any research yet, to the best of the authors’ knowledge, which is dedicated to address the sound emission due to flow around polygonal cylinders with a larger side number of
$N\gt 4$
. Inoue & Hatakeyama (Reference Inoue and Hatakeyama2002) conducted a direct numerical simulation on sound generation by a two-dimensional (2-D) circular cylinder placed in a uniform flow. Their findings revealed that the mechanism behind the dipole-type noise sources in the wake flow involves the alternative shedding of vortices from either side of the cylinder, leading to surface pressure fluctuations and acoustic waves which propagate outward into the surrounding fluid. Experiments conducted by Oguma, Yamagata & Fujisawa (Reference Oguma, Yamagata and Fujisawa2013) identified the sound sources located close to the flow separation points on the top and bottom sides of the cylinder, as well as in the near wake. Maryami et al. (Reference Maryami, Arcondoulis, Liu and Liu2023) further confirmed that pressure fluctuations imposed specifically on the circular cylinder shoulders play a critical role in the scattering of acoustic waves to the far field, thus contributing to the directivity signature of a dipole source. According to the aerodynamic noise mechanism of the circular cylinder, the alternate vortex-shedding behaviour behind the polygonal cylinder is likely to introduce similar aeroacoustic characteristics. Nevertheless, it is still of great interest to reveal how the polygonal cylinder modifies the characteristics and mechanism of sound generation, since it can effectively control the flow field and aerodynamic performance compared with the circular cylinder.
The present study aims to investigate the aeroacoustic characteristics and mechanism of flow around polygonal cylinders, given their relevance in engineering structures that may be situated near urban communities. We conduct a systematic experimental study on the far-field noise for polygonal cylinders of
$N\in {\mathbb{Z}}[3,12]$
in their principal (face and corner) orientations. Special attention is paid to the heptagonal cylinder in the corner orientation, denoted as 7C, where a distinctive dual-tone sound is clearly captured. Synchronised multi-field measurements of 2-D time-resolved particle image velocimetry (PIV), near-field pressure fluctuations and far-field noise are conducted to investigate the noise generation mechanism for this particular dual-tone phenomenon.
The experimental set-up is described in § 2. The characteristics of the far-field noise spectra are discussed in § 3, followed by detailed analyses of vortex-shedding pattern, shear-layer separation and reattachment and the corresponding relocation of sound sources for the heptagonal cylinder in § 4, where the mechanism behind the dual tone is explained. Finally, § 5 proposes a hypothesis on the mechanism of vortex-shedding mode switching, and § 6 concludes the main findings of this paper.
2. Experimental set-up
2.1. Wind tunnel and cylinder model
The experiments were carried out in a low-speed, closed-loop open-jet anechoic wind tunnel located at the Southern University of Science and Technology (SUSTech), as illustrated in figure 1. The wind tunnel features a rectangular test section measuring
$0.6\,{\textrm{m}} \,\times \, 0.55\,{\textrm{m}}$
, with a maximum free-stream velocity of
$U_{\infty }=70$
$\textrm {ms}^{-1}$
. The flow is introduced into the test section through seven layers of screens, a honeycomb structure, followed by a 7 : 1 contraction nozzle section, resulting in a turbulence intensity of less than 0.15 % at
$U_{\infty }=50$
$\textrm {ms}^{-1}$
. The inner dimensions of the anechoic chamber are
$3.8\,{\textrm{m}} \times 5.7\,{\textrm{m}} \times 3.0\,{\textrm{m}}$
, with a cutoff frequency of 100 Hz and a free-field radius of 1.9 m. Additional details on the anechoic wind tunnel can be found in Yang et al. (Reference Yang, Liu, Liu, Shen, Zhang, Wei, Liu and Xu2021). In the present study, the free-stream velocity varied from
$10$
$\textrm {ms}^{-1}$
to
$42$
$\textrm {ms}^{-1}$
, resulting in the corresponding Reynolds number
$\textit{Re} \approx 1.7\times 10^4$
to
$7 \times 10^4$
, based on the circum-circle diameter (outer diameter),
$D=25$
mm, of the cylinder.
A schematic illustration of the experimental set-up facilitating synchronised measurements of far-field sound, surface pressure and flow field. (
$a$
) Details of the cylinder mounting and measurement instrument arrangement. The PIV measurements of the two fields of view (FOVs) were conducted separately. (
$b$
) Top view of the midspan plane showing the two PIV FOVs and the arrangement of the eight surface microphones on the port side of the heptagonal cylinder surface. (
$c$
) Side view.

Ten polygonal cylinders of
$N=3$
–
$12$
, as well as a circular cylinder, were tested in this study. These cylinders have the identical outer diameter
$D=25$
mm and span
$L=550$
mm, resulting in an aspect ratio of
$L/D=22$
and a flow blockage ratio of 4.5 %. This very high aspect ratio ensures an approximately 2-D flow in the central region of the cylinder model. The cylinders were manufactured using photosensitive resin by 3-D printing, ensuring adequate structural strength and precision. A satisfactory surface finish was achieved, subject to the limitations of the 3-D printing precision, and was consistent across all cylinders. During the experiments, the cylinders were mounted vertically in the wind tunnel test section, having their bottom end firmly attached to a circular turning table, and their top end to a passive supporting circular plate. This arrangement allows for precise alignment of the cylinder with the flow direction. We focus on principal orientations of the polygonal cylinder only, which is a symmetric configuration, where either an edge or a flat surface normally faces the incoming flow (see figure 1
$a$
), referred to as corner or face orientation, respectively. The turning table and supporting plate are mounted flush with two end plates seamlessly attached to the nozzle exit, as shown in figure 1(
$a$
), which promotes a 2-D flow with negligible end effects (West & Apelt Reference West and Apelt1982).
2.2. Far-field noise and surface pressure measurements
Far-field noise measurements were performed on all the polygonal cylinders using a Brüel & Kjær (B&K) free-field type 4966 microphone, positioned at a distance of 1.5 m from the cylinder’s centre plane and 25 mm downstream of the cylinder axis. For synchronised measurement of far-field noise, surface pressure and PIV, the microphone distance was set closer at 1 m in order to prevent the contamination of reflected sound waves from the laser emitter, as shown in figure 1. The two microphone distances are equivalent to
$60D$
and
$40D$
, respectively, ensuring the far-field criterion (Henning et al. Reference Henning, Kaepernick, Ehrenfried, Koop and Dillmann2008; Geyer Reference Geyer2020). Acoustic signals were acquired using a National Instruments PXI-10,420 chassis equipped with a 24-bit NI PXI-4496 DAQ card, operating at a sampling frequency of 51.2 kHz for a duration of 20 s. Time-domain pressure fluctuations were converted into the frequency domain via the fast Fourier transform with Hamming windowing and 50 % overlap, following Welch’s method (Welch Reference Welch1967). The power spectral density (PSD),
$\varPhi _{pp}$
in
$\textrm {dB}\,\textrm {Hz}^{-1}$
, of the far-field noise is defined as
\begin{equation} \varPhi _{pp} = 10\log _{10} \left (\frac {\phi _{pp}}{p^{2}_{0}}\right )\!, \end{equation}
where
$\phi _{pp}$
is the PSD of acoustic pressure in
$\textrm {Pa}^2\,\textrm {Hz}^{-1}$
, and the reference pressure is taken as
$p_{0} = 20\,\unicode{x03BC}$
Pa.
Simultaneous measurements of surface pressure fluctuations, using Sonion 8002 surface microphones at the same sampling rate and for the same duration, were additionally conducted on the heptagonal cylinder in the corner orientation (i.e. 7C), inspired by the finding of the strong dual-tone phenomenon in far-field noise spectra. A total of eight surface microphones S1–S8 were evenly spaced along the cylinder surface in the midspan plane, with S1 located at the windward stagnation point and S8 at the leeward cylinder base (see figure 1
$b$
). These omnidirectional electret condenser microphones have a diameter of
$2.5$
mm and a circular sensing area with a
$0.8$
mm diameter. They have a sensitivity of
$21.1$
mV Pa−1 in the flat response frequency region (
$70{-}6000$
Hz). To mitigate the effect of high-frequency attenuation due to the small sensing area, these microphones were installed inside the cylinder with pinhole openings having the same size as the sensing area (Garcia-Sagrado & Hynes Reference Garcia-Sagrado and Hynes2012), as shown in figure 1(
$a$
). These surface microphones were calibrated prior to each test to correct for potential amplitude damping and phase delay caused by the pinhole. A 1/2 inch B&K free-field microphone type 4966, calibrated using a B&K type 4231 calibrator, served as the reference microphone. The PSD of surface pressure fluctuations,
$\varPhi _{pp,s}$
, was calculated following (2.1).
Repeatability tests showed that the measurement error for both surface pressure fluctuations and far-field acoustic pressure are within
$\pm 0.5$
dB. To assess their PSD uncertainties, we follow the convergence error estimation method for Fourier transform (Bendat & Piersol Reference Bendat and Piersol2011). Utilising 80 sub-blocks results in a random error of approximately 0.46 dB, calculated by
$10\log 10(1+1/\sqrt {N_b})$
, where
$N_b$
is the number of sub-blocks. In this study, the frequency resolution of both the
$\varPhi _{pp}$
and
$\varPhi _{pp,s}$
spectra is 4 Hz, and the frequency uncertainty at a 95 % confidence level is approximately
$\pm (2/\sqrt {3})\Delta f \approx \pm 4.6$
Hz (Porteous et al. Reference Porteous, Moreau and Doolan2017). This frequency uncertainty corresponds to an uncertainty of Strouhal number ranging from
$\pm 0.003$
to
$\pm 0.011$
within the free-stream velocity range tested, where the Strouhal number is defined as
$St = fD/U_{\infty }$
using the cylinder outer diameter as the characteristic length with
$f$
the frequency.
2.3. Time-resolved PIV measurements
To investigate the mechanism behind the dual-tone phenomenon, time-resolved planar PIV measurements were performed for the flow around the 7C cylinder, at three typical Reynolds numbers of
$\textit{Re}=2\times 10^4$
,
$2.5\times 10^4$
and
$4\times 10^4$
. The PIV measurements of two FOVs were performed separately, as shown in figure 1(
$b$
), both synchronised with the far-field and surface microphone measurements. The high-speed cameras utilised were Phantom VEO410L with 1280
$\times$
780 pixels, 12 bit,
$20 \ \unicode{x03BC} {\textrm{m}}$
pixel pitch and a maximum frame rate of
$5.2$
kHz. The large FOV1 in purple (see figure 1
$b$
), using a Nikon Macro-Nikkor
$100$
mm lens, captures the wake of the heptagonal cylinder in a region of
$3.5D\times 2D$
(trimmed according to illumination coverage), symmetric to the wake centreline and starting immediately from the rear cylinder surface. The small FOV2 in orange, with a Nikon Micro-Nikkor
$200$
mm lens and aiming to resolve the shear-layer separation dynamics, covers an area of
$1.5D\times 1D$
on the port side of the cylinder. Both FOVs were located in a plane with a 10 mm offset to the cylinder midspan, as shown in figure 1(
$a$
,
$c$
).
The PSD of the far-field noise,
$\varPhi _{pp}$
, of the polygonal cylinders with different side numbers, along with the circular cylinder for comparison. (
$a$
–
$c$
) Corner orientation, (
$d$
–
$f$
) face orientation; (
$a$
,
$d$
)
$\textit{Re} = 2\times 10^4$
, (
$b$
,
$e$
)
$\textit{Re} = 3\times 10^4$
, (
$c$
,
$f$
)
$\textit{Re} = 4\times 10^4$
. The insets in (
$c$
) and (
$f$
) are zoomed-in views of
$N=11$
and
$N=9$
, respectively.

The flow was seeded with Dioctyl sebacate particles having a mean diameter of
$1 \ \unicode{x03BC} {\textrm{m}}$
. Illumination was provided by a high-speed Photonics Nd:YLF laser with 50 mJ per pulse at 10 kHz. Imaging synchronisation was achieved using a LaVision programmable timing unit, which also triggers the acquisition of pressure data with the microphones. Image pairs were acquired at a sampling frequency of 1.6 kHz for a duration of 2 s, compromising the camera memory capacity and the highest frequency of interest. Over this duration, approximately 300 complete vortex-shedding cycles can be sampled for the higher-frequency shedding mode at
$\textit{Re}=2.5\times 10^4$
, for instance. The PIV
$\Delta t$
depends on the flow velocity and FOV size; for
$\textit{Re}=2.5\times 10^4$
, it was set as
$27.16\,\unicode{x03BC}$
s for FOV1 and
$9.7\,\unicode{x03BC}$
s for FOV2, within which the average particle displacement was approximately 7 pixels.
Particle images were processed by LaVision DaVis 10.1 software with an iterative multi-pass and a final interrogation window size of
$32\times 32$
pixels and
$75\,\%$
overlap. This gives spatial resolutions based on interrogation window size of
$2.8$
and
$1$
mm, respectively, for the two FOVs. In order to minimise the artefact from the usage of overlapped particle groups, when computing the velocity gradient, a central difference scheme using
$i\pm 2$
was adopted, instead of
$i\pm 1$
for the standard
$50\,\%$
interrogation window overlap. The calculated velocity gradient was then interpolated to restore the same grid spacing for the two velocity components.
3. Aeroacoustic characteristics
3.1. Far-field noise
3.1.1. Noise spectra and dual tones
Figure 2 presents the far-field noise spectra,
$\varPhi _{pp}$
, for all the tested cylinders in both corner (
$a$
–
$c$
) and face (
$d$
–
$f$
) orientations. The spectra of the circular cylinder clearly show a dominant tonal peak at
$St= 0.19$
, which is associated with the fundamental vortex shedding, consistent with many previous findings in the same
${\textit{Re}}$
regime, for example Norberg (Reference Norberg1986) and Zdravkovich (Reference Zdravkovich1997). The noise spectra of most polygonal cylinders exhibit this primary peak around
$St\approx 0.2$
. The Strouhal number of this fundamental tone at
$\textit{Re} = 2 \times 10^4$
, denoted as
${\textit{St}}_{0}$
, is listed in table 1 and compared with the experimental results reported in Xu et al. (Reference Xu, Zhang, Gan, Li and Zhou2017), showing a reasonable agreement.
As shown in figure 2(
$b$
,
$c$
), a clear exception occurs for
$N=7$
at
$\textit{Re}=3\times 10^4$
and
$4\times 10^4$
, where a secondary peak, which is even stronger than the primary peak in amplitude, can be observed at a higher frequency of
$St\approx 0.24$
, indicating a distinct dual-tone characteristic. In this study, these two tones are denoted as
${\textit{St}}_{01}$
and
${\textit{St}}_{02}$
, respectively, in which
${\textit{St}}_{02} \approx 1.4 \,{\textit{St}}_{01}$
. The emergence of this strong secondary peak is accompanied by a considerable attenuation of peak amplitude at
${\textit{St}}_{01}$
, compared with the other cylinders at the same condition. This secondary peak was also observed in the large eddy simulation (LES) of the 7C case at
$\textit{Re}=10^4$
by Masoudi et al. (Reference Masoudi, Sims-Williams and Gan2023), as well as of other polygonal cylinders of
$N=5$
–
$8$
at an incidence angle. However, due to the relatively short simulation duration of approximately
$30$
shedding cycles, the exact frequency of
${\textit{St}}_{02}$
in the velocity spectrum could not be determined accurately, and
${\textit{St}}_{02}$
was interpreted as a superharmonic of
${\textit{St}}_{01}$
. On the other hand, this secondary peak was not reported in the experiments by Xu et al. (Reference Xu, Zhang, Gan, Li and Zhou2017), who identified the dominant frequency via the fluctuating lift coefficient. Interestingly, the dual-tone phenomenon was not observed in the noise spectra of the same heptagonal cylinder but in the face orientation (i.e. 7F).
The frequencies of the fundamental tone,
${\textit{St}}_{0}$
, based on the circum-circle diameter
$D$
and the streamwise projection diameter
$D_p$
, respectively, obtained from the present study at
$\textit{Re} = 2 \times 10^4$
. Also listed are the experimental data from Xu et al. (Reference Xu, Zhang, Gan, Li and Zhou2017), where the
${\textit{St}}_0$
values based on
$D_p$
were obtained from the fluctuating lift coefficient at
$\textit{Re} = 2.1 \times 10^4$
. The case names of the polygonal cylinders consist of the side number
$N$
and orientation, i.e. C for corner orientation and F for face orientation.

Figure 2(
$c$
,
$f$
) reveals that dual peaks are also present in the spectra of the 11C and 9F cylinders, respectively, at
$\textit{Re}=4\times 10^4$
(see the insets). However, for these two cases, the secondary peak does not appear as distinct as that of 7C, with a much weaker amplitude than the primary one, leaving the primary peak remain dominant in the spectrum. For this reason, these two cases will not be further considered in this paper. Nevertheless, it is believed that the mechanism of the weaker secondary tone is similar to that of 7C, which will be discussed later in § 3.4.
Particular attention is then paid to the spectral behaviours of the heptagonal cylinder, which are essentially dependent on Reynolds number and are therefore anticipated to be within a transitional flow regime. Figure 3 provides a comparison of far-field noise spectra within a wide range of
${\textit{Re}}$
for corner and face orientations. The lower-frequency tone at
${\textit{St}}_{01}$
is labelled as
$f_{01}$
and the higher-frequency tone at
${\textit{St}}_{02}$
as
$f_{02}$
, which begins to emerge from
$\textit{Re}=3\times 10^4$
. However, this secondary tone
$f_{02}$
appears for the corner orientation only, as shown in figure 3(
$a$
), and is never observed for the 7F case in the same
${\textit{Re}}$
range (see figure 3
$b$
). A very weak peak at
$St \approx 0.095$
can be seen at the lowest free-stream velocity (
$\textit{Re} = 2\times 10^4$
) in both orientations, as well as in figure 2(
$a$
,
$d$
), which is likely due to the background noise from the wind tunnel cooling system resonance, and does not affect the following analysis since it is well separated from both tones of
$f_{01}$
and
$f_{02}$
in the spectrum. A cross-comparison of figures 3(
$a$
) and 3(
$b$
) reveals that both tonal peaks in 7C are significantly lower than the single, dominant tone
$f_{0}$
in 7F.
The PSD of the far-field noise,
$\varPhi _{pp}$
, for the heptagonal cylinder in corner and face orientations within the
${\textit{Re}}$
range of
$2\times 10^4\leqslant Re\leqslant 7\times 10^4$
. Panels show (
$a$
) 7C and (
$b$
) 7F.

(
$a$
) Contours of far-field noise PSD,
$\varPhi _{pp}$
, with respect to
${\textit{Re}}$
and
$St$
for the 7C case. Here, RI, RII and RIII represent the three
${\textit{Re}}$
regimes according to the relative amplitudes of the two tones. (
$b$
) The frequency variation of the two tones,
$f_{01}$
and
$f_{02}$
, with
${\textit{Re}}$
for 7C.

The relative amplitudes of the two tones,
$\varPhi _{pp}({f_{01})}$
and
$\varPhi _{pp}({f_{02})}$
, as well as their bandwidths, are more clearly shown in the contour plot of figure 4(
$a$
), in which three flow regimes can be categorised according to the dependence of tonal characteristics on the Reynolds number. Regime I (RI) refers to the range of
$\textit{Re} \lesssim 2.5\times 10^4$
, in which tone
$f_{01}$
dominates and tone
$f_{02}$
emerges in the spectrum at the end of this regime. Regime II (RII) is the
${\textit{Re}}$
range (
$2.5\times 10^4 \lesssim Re \lesssim 6\times 10^4$
) where
$f_{02}$
begins to dominate the noise spectrum in both amplitude and bandwidth, particularly at higher
${\textit{Re}}$
above
$4.5\times 10^4$
. In regime III (RIII) of
$\textit{Re} \gtrsim 6\times 10^4$
, however, the
$f_{02}$
tone rapidly attenuates, allowing
$f_{01}$
reverse as the dominant tone once again. A further justification for the division of these
${\textit{Re}}$
regimes will be provided in § 3.4. This transition of the dual tones in amplitude with
${\textit{Re}}$
is prominently different from the circular cylinder, for which a single, dominant peak persists in the spectrum throughout the subcritical regime of
$\textit{Re} \lesssim 10^5$
(Norberg Reference Norberg1994; Zdravkovich Reference Zdravkovich1997; Maryami, Arcondoulis & Liu Reference Maryami, Arcondoulis and Liu2024).
The frequencies of the two tones,
${\textit{St}}_{01}$
and
${\textit{St}}_{02}$
, in figure 4(
$a$
) are further quantified and presented in figure 4(
$b$
) with varying Reynolds number. The uncertainties associated with
${\textit{St}}_{01}$
and
${\textit{St}}_{02}$
, as discussed in § 2.2, are illustrated as the error bar. As shown in figure 4(
$b$
), the average values of the two tonal frequencies over the
${\textit{Re}}$
range are approximately
$\overline {St}_{01} \approx 0.17$
and
$\overline {St}_{02} \approx 0.24$
, respectively. At the end of regime I,
${\textit{St}}_{01}$
slightly reduces from
$0.18$
to
$0.17$
as the
$f_{02}$
tone emerges. The two tonal frequencies then almost stabilise around
$St\approx 0.17$
and
$0.24$
, respectively, over the most of regime II. As
${\textit{St}}_{02}$
gradually decreases in regime III, the
$f_{01}$
tone returns to a relatively stable frequency of
${\textit{St}}_{01} \approx 0.175$
.
3.1.2. Overall sound pressure level
By integrating the far-field PSD over the frequency range of 60–20 000 Hz, the overall sound pressure level (OASPL) can be calculated as follows:
\begin{equation} {\textrm{OASPL}} = 10\log _{10}\left (\int _{60}^{20\,000} \phi _{pp}/p_0^{2} \; {\textrm{d}}f\right )\!. \end{equation}
This lower bound of 60 Hz ensures that the main spectral peaks are well included in the integration, while the weak peak from background noise, as discussed in figure 3, is excluded. For instance, for the noise spectra at
$\textit{Re} = 2\times 10^4$
in figure 2(
$a$
,
$d$
), the lower bound frequency corresponds to
$St=0.125$
, which is above the frequencies of the weak peaks yet well below those of the primary tones.
Figure 5(
$a$
) shows the far-field OASPL of the 7C and circular cylinders, as well as the total sound power of the
$f_{01}$
and
$f_{02}$
tones, which is obtained by integrating over the tonal peak with the frequency bounds defined by those with 15 dB below the peak PSD level (Geyer Reference Geyer2020; Bai et al. Reference Bai, Lu, Wei, Yang and Liu2021). It is well known that the OASPL of aerodynamic noise from flow around a cylinder is dominated by the primary tone due to vortex shedding (Choi et al. Reference Choi, Jeon and Kim2008). Similarly, the OASPL of 7C appears to be dominated by the stronger tone between
$f_{01}$
and
$f_{02}$
over the Reynolds number range. The relative OASPLs of the two tones across the three regimes appear consistent with the relative amplitudes of their tonal peaks,
$\varPhi _{pp}({f_{01})}$
and
$\varPhi _{pp}({f_{02})}$
, as shown in figure 4(
$a$
).
(
$a$
) The OASPL variation of 7C with
${\textit{Re}}$
as well as those of
$f_{01}$
and
$f_{02}$
separately, in comparison with the circular cylinder. (
$b$
–
$d$
) Value of
$\Delta$
OASPL of the polygonal cylinders against the circular cylinder at three Reynolds numbers: (
$b$
)
$\textit{Re}=2\times 10^4$
(RI), (
$c$
)
$\textit{Re}=4\times 10^4$
(central RII) and (
$d$
)
$\textit{Re}= 7\times 10^4$
(RIII).

Figure 5(
$a$
) shows that the OASPL of both the 7C and circular cylinders generally follows the 6th power law with Mach number,
$ \textit{Ma}$
, of the incoming flow, which is typical of the dipole sound source (Curle Reference Curle1955; Phillips Reference Phillips1956; Duan & Wang Reference Duan and Wang2021). Therefore, it can be inferred that, despite having two tones, the 7C wake flow essentially features vortex shedding from both sides of the cylinder, leading to symmetric surface pressure fluctuations (Inoue & Hatakeyama Reference Inoue and Hatakeyama2002), which, in turn, generate dipole sound waves propagating outward. A remarkable reduction of OASPL by 2.8–9.8 dB is achieved by 7C relative to the circular cylinder. At low Reynolds numbers in RI, the
$f_{02}$
tone is hardly noticeable or just emerging in the spectrum, and the OASPL reduction is relatively small at approximately 3 dB. In RII with increasing
${\textit{Re}}$
,
$f_{02}$
quickly takes dominance over
$f_{01}$
in amplitude, resulting in a significant OASPL reduction of 4.3–8.5 dB. As
${\textit{Re}}$
further increases in RIII, despite
$f_{02}$
attenuating in amplitude and
$f_{01}$
recovering as the dominant tone, a maximum OASPL reduction of almost 10 dB is achieved.
Figure 5(
$b$
–
$d$
) presents the OASPL for the polygonal cylinders of
$N\in {\mathbb{Z}}[3,12]$
in corner and face orientations, for three Reynolds numbers, one in each regime of RI–RIII defined based on 7C. The OASPL values are benchmarked against those of the circular cylinder, and hence
$\Delta {\textrm{OASPL}}\lt 0$
indicates a noise reduction with respect to the circular cylinder. Apparently, the triangular cylinder in corner orientation (3C) has the highest noise reduction through all
${\textit{Re}}$
tested, which can be attributed to its most streamlined geometry of the windward surface. This is in line with its lowest root mean square of the fluctuating lift coefficient compared with other polygonal cylinders reported in Xu et al. (Reference Xu, Zhang, Gan, Li and Zhou2017). A rigorous link between shape streamlining and noise reduction could be investigated in future work, addressing the effects of shear-layer interaction and the vorticity-induced body force through detailed near-wall flow-field data. The marginally negative
$\Delta {\textrm{OASPL}}$
for the square cylinder in the corner orientation (4C) can also be ascribed to this reason. As the side number
$N$
increase, the windward surface geometry becomes more blunt (in the corner orientation), and this noise reduction effect rapidly diminishes. Notwithstanding, the distinguishable noise reduction of 7C can be observed, with the strongest effect in RIII, as has been shown in figure 5(
$a$
).
Combining the
$ \textit{Ma}^6$
scaling law of the dipole sound source with the prominent noise reduction effect observed in the 7C cylinder, it can be hypothesised that the
$f_{02}$
tone is a manifestation of similar vortex-shedding behaviours leading to the
$f_{01}$
tone. Nevertheless, as the Reynolds number increases, the emergence and subsequently the dominance of the
$f_{02}$
tone results in the formation of weaker vortices shed into the wake, thereby weakening the far-field acoustic radiation, as will be discussed in § 4.3.
For
$N\gt 4$
, apart from 7C, only three cases show a clear negative
$\Delta {\textrm{OASPL}}$
. The noise reductions of
$N=9$
in the face orientation (9F) and
$N=11$
in the corner orientation (11C) in figure 5(
$c$
,
$d$
), i.e.
$\textit{Re}\geqslant 4\times 10^4$
, are attributed to the same dual-tone phenomenon as for 7C, highlighted in figure 2(
$c$
,
$f$
). The obvious noise reduction of
$N=10$
in face orientation (10F) at
$\textit{Re} = 2\times 10^4$
(see figure 5
$b$
) is due to its particularly low PSD level of the primary tone at this
${\textit{Re}}$
, which rapidly rises and surpasses that of the circular cylinder as
${\textit{Re}}$
increases, resulting in positive
$\Delta {\textrm{OASPL}}$
in figure 5(
$c$
,
$d$
). To determine the exact physical mechanism of 10F requires a lower
${\textit{Re}}$
testing range, which is left for future study.
3.2. Surface pressure fluctuation
(
$a$
) The PSD of surface pressure fluctuations,
$\varPhi _{pp,s}$
, on 7C at
$\textit{Re} = 2.5 \times 10^4$
, where the surface microphones S2, S4 and S6 are positioned at the centre of their respective side surfaces. (
$b$
–
$d$
) Contours of surface pressure PSD,
$\varPhi _{pp,s}$
, with respect to
${\textit{Re}}$
and
$St$
, for S2, S4 and S6, respectively.

The mechanism of the single primary tone of
$\varPhi _{pp}$
in most cases shown in figure 2 is plausibly similar to that of the circular cylinder, in that the dominant vortex shedding from the cylinder leads to strong surface pressure fluctuations and acoustic waves radiated to the far field, which is well established (Inoue & Hatakeyama Reference Inoue and Hatakeyama2002; Maryami et al. Reference Maryami, Arcondoulis, Liu and Liu2023). The distinctive dual-tone phenomenon in flow around 7C is further characterised by surface pressure spectra,
$\varPhi _{pp,s}$
(
$\textrm {dB}\,\textrm {Hz}^{-1}$
), calculated following (2.1). Figure 6(
$a$
) presents the PSD of surface pressure fluctuations measured by surface microphones S2, S4 and S6, located at the centre of the three side surfaces on the port side of 7C, respectively. The same dual-tone behaviour,
$f_{01}$
and
$f_{02}$
, can be clearly observed in all three spectra, as with those in figure 3(
$a$
). This feature of
$\varPhi _{pp,s}$
observed from the surface pressure fluctuations confirms the characteristics of the particular vortex-shedding pattern, which is responsible for generating the dual-tone acoustic sources through imposing corresponding pressure fluctuations on the cylinder surface, consistent with Curle’s analogy (Curle Reference Curle1955).
The upstream influence of vortex shedding on pressure fluctuations on the cylinder surface, known as the feedback signal, depends on the distance between the cylinder surface and the vortex formation region (Maryami & Liu Reference Maryami and Liu2024; Maryami et al. Reference Maryami, Arcondoulis and Liu2024). As this distance increases, the intensity of the feedback signal diminishes due to increased energy dissipation. As a result, the signal intensity at the S2 microphone, which is the most upstream and the furthest from the shear-layer separation point among the three, is obviously the lowest. Evidence will be provided in § 4. It is noteworthy that S4 and S6 exhibit remarkably similar levels of
$\varPhi _{pp,s}$
at the two tonal peaks. Moreover,
$\varPhi _{pp,s}(f_{02})$
at S6 is marginally higher than that at S4, indicating that the acoustic source associated with
$f_{02}$
is further downstream, i.e. closer to the S6 microphone.
Figure 6(
$b$
–
$d$
) displays the
$\varPhi _{pp,s}$
contours at S2, S4 and S6, with the reference
${\textit{Re}}$
regimes RI, RII and RIII labelled according to the far-field noise results discussed in § 3.1. Similar to figure 4(
$a$
), double bands of the dual tones,
$f_{01}$
and
$f_{02}$
, are evident. In RII, the appearance of
$f_{02}$
is accompanied by the subtle decrease of
$f_{01}$
in amplitude, while in RI,
$f_{01}$
is almost the only tone observable. However, comparing with figure 4(
$a$
), while
$f_{02}$
arises at a similar Reynolds number, i.e. at the start of RII at
$\textit{Re} \approx 2.5\times 10^4$
, it fades away at higher
${\textit{Re}}$
values. In RIII,
$f_{02}$
further attenuates and
$f_{01}$
recovers as the only primary tone. This could be attributed to the finite pinhole size to accommodate the surface microphones at the cylinder edges (i.e. S1, S3, S5 and S7, not shown), which inevitably slightly flattens the edge locally. Nevertheless, in RII the acoustic signal of the
$f_{02}$
tone remains prominent to allow for the following analysis.
3.3. Wavelet transform analysis
In this section, continuous wavelet transform (CWT) is applied to the pressure data from both far-field and surface microphone measurements to evaluate the possible temporal behaviour of their frequency spectra, which is defined as
where the wavelet coefficient
$W(a,\tau )$
is the CWT of the pressure signal
$p^{\prime}(t)$
,
$a$
is the scale dilation parameter and
$\tau$
is the time delay. Also,
$\psi (t)$
is the mother wavelet basis function, a continuous function in both time and frequency domains,
$\psi ^{*}[(t-\tau )/a]$
is the complex conjugate of the dilated and translated
$\psi (t)$
and
$C^{-1/2}_{\psi }$
is a constant that takes the mean value of
$\psi (t)$
. As a common mother wavelet, a Morlet wavelet is used for
$\psi$
(Farge Reference Farge1992; Lysenko, Ertesvåg & Rian Reference Lysenko, Ertesvåg and Rian2014; Maryami et al. Reference Maryami, Arcondoulis and Liu2024), and the parameters are calculated automatically according to the window size, which are insensitive to the results presented below.
Wavelet scalograms of far-field noise for the heptagonal cylinder in corner and face orientations at
$\textit{Re} = 2.5\times 10^4$
. Panels show (
$a$
) 7C, (
$b$
) 7F.

Figure 7 presents the CWT scalograms of far-field noise for the heptagonal cylinder in principal orientations at
$\textit{Re} = 2.5 \times 10^4$
, i.e. the lower bound of RII. An arbitrarily chosen time window of
$3$
s is presented for both orientations, where a clear difference in the frequency band of strong amplitudes can be noticed. In figure 7(
$b$
), a narrow band is centred at
$St \approx 0.16$
, corresponding to the single primary tone
$f_{0}$
in figure 3(
$b$
). However, a close inspection of figure 7(
$a$
) reveals two sub-bands centred at
${\textit{St}}_{01}\approx 0.17$
and
${\textit{St}}_{02}\approx 0.24$
, matching the two tones
$f_{01}$
and
$f_{02}$
in figures 3(
$a$
) and 4. Figure 7(
$a$
) also suggests that these two tones do not occur simultaneously; instead, they alternate over time and switch between each other in a manner that appears stochastic in nature.
Synchronised wavelet scalograms for 7C at
$\textit{Re} = 2.5 \times 10^4$
. (
$a$
) Far-field noise, (
$b$
–
$d$
) surface pressure fluctuations of S2, S4 and S6, respectively. The frequencies of
${\textit{St}}_{01}=0.17$
and
${\textit{St}}_{02}=0.24$
are marked by dashed lines.

The results in figure 7(
$a$
) are represented in figure 8(
$a$
) within a shorter time window of 1 s, along with CWT scalograms from synchronous surface pressure fluctuations for exactly the same time window shown in figure 8(
$b$
–
$d$
) for S2, S4 and S6, respectively. The time delay between the surface pressure and far-field noise signals was estimated through a cross-correlation analysis, and was then used to synchronise the temporal signals prior to the CWT analysis. The switching of strong-amplitude bands for the surface pressure fluctuations at different positions approximately synchronise among one another. However, the strong-amplitude bands of the far-field noise in figure 8(
$a$
) appear markedly different to those of the surface pressure signals. For instance, while the band is centred at
${\textit{St}}_{01}$
for the three surface pressure signals between t = 4.7 and 4.75 s, the band is centred at
${\textit{St}}_{02}$
for the far-field noise. This could be attributed to 3-D flow effects, leading to a finite spanwise correlation length, which, in turn, results in the surface pressure signals at this spanwise location differing from the global acoustic signal captured in the far-field at some instants. Since the S4 and S6 microphones are located within the (separated) shear layer, evidence to be provided in § 4, their signals are affected by stronger turbulence, which explains the occurrence of
$W$
spikes at higher frequencies, i.e.
$St\gt 0.3$
, more at S6 than at S4.
3.4. Time distribution of dual tones
The
${\textit{Re}}$
dependence of the far-field noise CWT results for 7C is analysed for a longer time duration of 4 s, as shown in figure 9. We select three typical
${\textit{Re}}$
cases, which are well within RI, RII and RIII (see figure 4
a). In all three cases, strong-amplitude bands switch between the two main tones
${\textit{St}}_{01}$
and
${\textit{St}}_{02}$
stochastically. However, the total duration covered by each tone appears to differ significantly. In RI (9
$a$
), the
${\textit{St}}_{01}$
band appears to dominate, with
${\textit{St}}_{02}$
only occurring occasionally. In RII (9
$b$
), the
${\textit{St}}_{02}$
band occupies the entire time window almost exclusively, while in RIII (9
$c$
) dominance shifts back to
${\textit{St}}_{01}$
, although not as strongly as that in figure 9(
$a$
), with strong amplitudes also spending more time in intermediate frequencies between
${\textit{St}}_{01}$
and
${\textit{St}}_{02}$
. Generally, as
${\textit{Re}}$
increases across RI–RIII with more turbulent flow, the switching between the two tones appears more frequently and stochastically.
Wavelet scalograms of far-field noise for 7C at (
$a$
)
$\textit{Re}=2.2\times 10^4$
(RI), (
$b$
)
$\textit{Re}=5\times 10^4$
(RII) and (
$c$
)
$\textit{Re}=6.7\times 10^4$
(RIII). The frequencies of
${\textit{St}}_{01}=0.17$
and
${\textit{St}}_{02}=0.24$
are marked by dashed lines.

The percentage of time spent in each tone relative to the total sampling duration can be quantified by employing an empirical threshold
$|W|\geqslant 0.86 |W|_{\textit{max}}$
followed by a time counting, subject to energy leakage in CWT (Mallat Reference Mallat1989) and the loss of time resolution for low-frequency signals (Daubechies Reference Daubechies1992). The outcome of this statistical quantification is shown in figure 10 using the CWT results of far-field noise only, for the entire duration of 20 s, where an interesting pattern is displayed in terms of the time percentages between the dual tones
$f_{01}$
and
$f_{02}$
. The set threshold gives rise to a total
$\kappa =95\,\%$
of sample time occupied by these two tones, within which each tone manifests a near Gaussian distribution over
${\textit{Re}}$
. Furthermore, the time percentages of
$f_{01}$
and
$f_{02}$
are nearly symmetrically distributed, reflecting the rather binary switch of the two tones. The probability of occurrence can be fitted with the empirical Gaussian distributions
\begin{equation} \mathcal{P}\left (f_{01},f_{02}\right ) = \frac {\kappa }{2} \pm \kappa \Bigg \{\exp \left [-\left ( \frac {Re - Re_T}{\sigma } \right )^2\right ]-\frac {1}{2}\Bigg \}, \end{equation}
where the standard deviation
$\sigma =1.2\times 10^4$
, the centre of the transition range
$\textit{Re}_T=4.3\times 10^4$
and ‘
$+$
’ sign for
$\mathcal{P}(f_{01})$
and ‘
$-$
’ sign for
$\mathcal{P}(f_{02})$
. The quality of the empirical function fitness is measured by the goodness of fit,
$R^2$
(Harrell Reference Harrell2015), whose values are 0.898 and 0.903 for
$f_{01}$
and
$f_{02}$
modes, respectively, suggesting a sufficiently good fit.
As shown in figure 10, for
$\textit{Re} \lt 2 \times 10^4$
, the probability of
$f_{02}$
stays low, but not null, consistent with the observation in figure 9(a), which also explains the small hump attached to
$f_{01}$
in figure 2(a). The predominance of
$f_{01}$
is overtaken by
$f_{02}$
at
$\textit{Re}\approx 2.7 \times 10^4$
, which then maximises at
$\textit{Re}_T$
, where the occurrence of
$f_{01}$
becomes most sparse. At a further higher
$\textit{Re} \approx 6\times 10^4$
, the probability dominance switches back, in line with the PSD and CWT results in figures 4(
$a$
) and 9, as well as the bimodal distribution of OASPL in figure 5. Nevertheless, the sum of the time percentages of
$f_{01}$
and
$f_{02}$
remains around 95 %, particularly in RI and RII, yet it drops noticeably below 95 % in RIII because more time is spent in frequencies between the dual tones (see figure 9
$c$
). The two critical Reynolds numbers, where
$\mathcal{P}(f_{01})\approx \mathcal{P}(f_{02})$
, are used to redefine the division of RI, RII and RIII, resulting in a minor correction to the RI–RII boundary from
$\textit{Re}\approx 2.5 \times 10^4$
in § 3.1.1 to approximately
$2.7 \times 10^4$
. Additionally, the approximated Gaussian distribution becomes slightly asymmetric at the two ends, i.e. the decaying rate in the probability of
$f_{02}$
at
$\textit{Re}\gt 7\times 10^4$
is slower than its rising rate in the low-
${\textit{Re}}$
region.
Variation of the time percentages corresponding to the
$f_{01}$
and
$f_{02}$
tones of 7C with Reynolds number. Lines show -
$\boldsymbol{\cdot}$
-:
$\mathcal{P}(f_{01},f_{02})$
in (3.3),
$\hbox{-}\hbox{-}$
:
$95\,\%$
confidence level of prediction bounds.

From the competing behaviour of the two main tones shown in figure 10, it is reasonable to speculate that the dual-tone transition phenomenon for 7C occurs only for
$10^4\lesssim Re \lesssim 10^5$
, outside of which, only a single tone
$f_{01}$
persists, and the chance for the secondary tone
$f_{02}$
to occur would be very low, if not zero. The results at the two very high
${\textit{Re}}$
of
$8\times 10^4$
and
$9\times 10^4$
were measured later to examine the empirical (3.3), and they align reasonably well with the fitted Gaussian distribution.
The two other cases highlighted in figure 2(
$c$
,
$f$
) and discussed in figure 5(
$c$
,
$d$
), viz.,
$N=9$
in the face orientation (9F) and
$N=11$
in the corner orientation (11C), could also be characterised by this dual-tone transition phenomenon. That is, distributions of
$\mathcal{P}(f_{01})$
and
$\mathcal{P}(f_{02})$
in the tested
${\textit{Re}}$
range seem describable by an equation similar to (3.3), but with a larger transition centre,
$\textit{Re}_T$
, closer to the maximum
${\textit{Re}}$
tested (figure not shown for brevity). This speculation will be validated in future work by using a higher-
${\textit{Re}}$
testing range.
4. Wake and near-wall flow fields of 7C
4.1. Time-averaged flow field
To unveil the formation mechanism of the dual-tone phenomenon in 7C flow, time-resolved planar PIV measurements were performed in the cylinder wake, as well as on the port side of the cylinder, at three representative Reynolds numbers, i.e.
$\textit{Re}=2\times 10^4$
(RI),
$2.5\times 10^4$
(RI–RII transitional regime) and
$\textit{Re}=4\times 10^4$
(central RII); see figures 4 and 10. Figure 11(
$a$
,
$b$
) presents the time-averaged velocity contours at
$\textit{Re}=2.5\times 10^4$
for the streamwise
$\overline {u}$
and transverse
$\overline {v}$
components, respectively, which resemble the characteristics of time-averaged velocity field behind the circular and other polygonal cylinders (Wang et al. Reference Wang, Xu, Gan, Zhang and Zhou2019; Maryami et al. Reference Maryami, Arcondoulis and Liu2024). The presence of the recirculation bubble is evident, marked by the red curve, outside which is the velocity acceleration region.
Contours of time-averaged flow velocity at
$\textit{Re} = 2.5\times 10^4$
for the (
$a$
) streamwise
$\overline {u}/U_{\infty }$
and (
$b$
) transverse
$\overline {v}/U_{\infty }$
components. Red solid curves mark the recirculation bubble region; the asterisk marks the strongest reverse flow location in the wake flow field,
$(x_r,0)$
. (
$c$
) Variation of
$\overline {u}$
with
$x$
along the centreline
$y=0$
, (
$d$
) variation of
$\overline {v}$
with
$y$
along
$x=x_r$
, marked by the red dashed line in (
$b$
).

As
${\textit{Re}}$
increases, although the overall distributions of
$\overline {u}$
and
$\overline {v}$
remain similar, the size of the recirculation bubble exhibits a discernible change in figure 11(
$c$
,
$d$
). The variation of
$\overline {u}$
along the wake centreline
$y=0$
shown in figure 11(
$c$
) illustrates that, as
${\textit{Re}}$
increases from RI to central RII, the recirculation bubble elongates, which is reflected by the downstream shift of the wake saddle position (i.e.
$\overline {u}=0$
). Alongside this change is the shrinkage of the bubble width, which can be inferred in figure 11(
$d$
) by the variation of
$\overline {v}$
along the transverse direction through
$(x_r,0)$
, the position of
$\overline {u}_{{min}}$
in the wake flow. In particular,
$|\overline {v}|_{\textit{max}}$
and the
$|y|$
-coordinate where this occurs, both dwindle as
${\textit{Re}}$
increases.
4.2. Velocity spectra and frequency segregation
The time series of transverse velocity,
$v(t)$
, across the wake are investigated at four sampling locations P1–P4 at
$(x/D,y/D)=(2,[0,-0.5,-1, 1])$
, which are marked in figure 12(
$a$
), with the background shown as vorticity field (
$\omega =\partial v/\partial x-\partial u/\partial y$
), in time-mean and non-dimensional form. In this and the following sections, we focus on the case at
$\textit{Re}=2.5\times 10^4$
to investigate the mechanism of the dual-tone phenomenon. According to figure 10, at this Reynolds number, the two main tones
$f_{01}$
and
$f_{02}$
have a similar probability, neither significantly dominating over the other.
(
$a$
) Time-averaged vorticity field at
$\textit{Re} = 2.5 \times 10^4$
. (
$b$
) The PSD of
$v^{\prime}(t)$
,
$\varPhi _{vv} = 10\log _{10}\phi _{vv}{}(\text{dB Hz}^{-1}$
, ref. 1 m
$\text{s}^{-1}$
), for the four sampling points in the wake flow along
$x/D=2$
at P1 (
$y/D=0$
, the wake centreline), P2 (
$y/D=-0.5$
, approximately inside the time-averaged shear-layer region), P3 (
$y/D=-1$
, outside the shear layer) and P4 (
$y/D=1$
, symmetrical with P3 to the wake centreline). (
$c$
,
$d$
) The CWT scalograms for
$v^{\prime}(t)$
acquired at P3 and P4, respectively. (
$e$
) Time segments of occurrence of the
$f_{01}$
and
$f_{02}$
modes at P1 based on the thresholded CWT. The time intervals
$\Delta t_1 = 1$
–
$1.4$
s and
$\Delta t_2 = 0.7$
–
$0.9$
s are for the cross-correlation analysis of the two modes in § 4.8.

Figure 12(
$b$
) presents the PSD of transverse velocity fluctuation
$v^{\prime}(t)$
acquired at the four locations, where
$\varPhi _{vv}$
is computed according to (2.1) with a reference velocity of 1
$\textrm {ms}^{-1}$
. The overall spectral behaviour shows that as the wake centreline is approached, the spectral level increases, i.e.
$\varPhi _{vv}(\text{P}4)\approx \varPhi _{vv}(\text{P}3)\ll \varPhi _{vv}(\text{P}2)\lesssim \varPhi _{vv}(\text{P}1)$
, and the approximate levels of P3 and P4 indicate an almost symmetric flow field to the wake centreline. This is in line with the feature of turbulence field in the wake behind general polygonal cylinders in principal orientations (Wang et al. Reference Wang, Xu, Gan, Zhang and Zhou2019). At each sampling point, the emergence of two tonal peaks is evident, which are at
${\textit{St}}_{01}\approx 0.17$
and
${\textit{St}}_{02}\approx 0.24$
, consistent with the spectra of far-field noise and surface pressure fluctuations, shown in figures 3(
$a$
) and 6(
$a$
), respectively. Since it is widely established that the velocity spectrum signifies a vortex-shedding pattern, this observation suggests that two modes of vortex shedding occur, one mode at a distinctive frequency, both temporarily stable. Following the analytical model of Gloerfelt et al. (Reference Gloerfelt, Pérot, Bailly and Juvé2005), the far-field dipole sound generated by the cylinder was attributed to an equivalent source due to the scattering of quadrupole volume sources in the wake by the cylinder’s surface. This further confirms that these two vortex-shedding modes are the origin of the dual tones observed in the far-field noise and surface pressure spectra.
Figure 12(
$c$
,
$d$
) shows the CWT scalograms for the
$v^{\prime}(t)$
signal acquired at P3 and P4, respectively. The temporary distributions of the two tones are almost identical between P3 and P4, which indicates that the same vortex-shedding mode occurs on both sides of the cylinder simultaneously and further confirms the symmetry of the wake field. A further examination of the velocity spectra reveals that
$\varPhi _{vv}(f_{01})\gt \varPhi _{vv}(f_{02})$
at all four sampling locations, consistent with the PSD of surface pressure fluctuations,
$\varPhi _{pp,s}(f_{01})\gt \varPhi _{pp,s}(f_{02})$
, as shown in figure 6(
$a$
). This can be attributed to two possible factors: firstly, the vortex-shedding mode of the
$f_{01}$
tone may occur more frequently than the mode of
$f_{02}$
over the PIV sampling duration of 2 s; secondly, energy of the
$f_{01}$
mode can itself be stronger than that of
$f_{02}$
.
In order to isolate the two factors, we first address the probability distributions of
$f_{01}$
and
$f_{02}$
through a frequency segregation method, which follows § 3.4 using CWT analysis. The probability of occurrence for each of the two modes can be obtained by properly thresholding the wavelet coefficient
$W$
in the same way as described in § 3.4, i.e.
$|W|\geqslant 0.86 |W|_{\textit{max}}$
. After applying the wavelet transform to the
$v^{\prime}(t)$
signal at P1 to extract time-resolved wavelet coefficients at the two dominant shedding frequencies (
$St= 0.17$
and
$0.24$
), we normalise the amplitudes of
$|W|$
globally, and assign each snapshot to the mode whose normalised wavelet amplitude exceeds a preset threshold of 0.86.
The time segments of occurrence of the two modes are identified in figure 12(
$e$
), from which the probabilities of
$f_{01}$
and
$f_{02}$
can be estimated as
$0.6$
and
$0.26$
, respectively, hence validating the first factor. These time percentages from PIV snapshots are largely aligned with those calculated from (3.3), i.e.
$\mathcal{P}(f_{01})\approx 0.55\pm 0.21$
and
$\mathcal{P}(f_{02})\approx 0.4\pm 0.21$
, fitted from far-field noise results. The discrepancy is ascribed to the much shorter sampling duration of PIV (2 s), compared with that of far-field noise (20 s), and the stochastic nature of shedding mode switching.
4.3. Spectrally averaged flow field
Following the frequency segregation method described in figure 12, PIV snapshots can be categorised according to the temporarily dominant vortex-shedding mode,
$f_{01}$
or
$f_{02}$
, from which time-averaged velocity fields
$\overline {u}(f_{01},f_{02})$
,
$\overline {v}(f_{01},f_{02})$
can be calculated, along with the corresponding turbulence fields, subjected to different statistical convergence due to unequal number of snapshots available for each mode. Such a mode-dependent flow field is referred to as a spectrally averaged field in this paper. Contours for
$\overline {u}(f_{01},f_{02})$
and
$\overline {v}(f_{01},f_{02})$
are presented in figures 13(
$a$
) and 13(
$b$
), which share patterns similar to those in figure 11(
$a$
,
$b$
). However, it is not difficult to find from the recirculation bubble shapes and the
$\overline {v}$
contour distribution in (b) that mode
$f_{01}$
has a larger wake than mode
$f_{02}$
. We now focus more on turbulence quantities, as the mode-dependent variations are more clearly manifested there.
Figure 13(
$c$
–
$e$
) shows the Reynolds stress contours with those calculated from modes
$f_{01}$
and
$f_{02}$
placed next to each other for a clearer comparison. Here,
$u'(f_{01})=u(f_{01})-\overline {u}(f_{01})$
and
$v'(f_{01})=v(f_{01})-\overline {v}(f_{01})$
, where
$u(f_{01})$
and
$v(f_{01})$
are streamwise and transverse velocity components, respectively, for the snapshots where
$f_{01}$
dominates (see figure 12
$e$
). Same notations apply to the snapshots dominated by the
$f_{02}$
mode.
In figure 13(
$c$
), the contour plot of streamwise normal Reynolds stress
$\overline {u'u'}/U_{\infty }^2$
reveals distinct differences in wake width between the two modes
$f_{01}$
and
$f_{02}$
. The half-wake width
$D_w/2$
measured at the maximum
$\overline {u'u'}$
is approximately
$0.49D$
for the
$f_{01}$
mode, while for
$f_{02}$
it is approximately
$0.32D$
, significantly narrower, indicating an approximately
$35\,\%$
reduction. This pronounced narrowing of the wake in mode
$f_{02}$
aligns with the subsequent theoretical analysis that a smaller wake width is associated with higher vortex-shedding frequencies, thereby confirming that the high-frequency acoustic tone originates from a wake flow with reduced lateral extent.
Time-averaged flow velocities and turbulence quantities associated with the
$f_{01}$
and
$f_{02}$
modes at
$\textit{Re}=2.5\times 10^4$
. Panels show (
$a$
)
$\overline {u}/U_{\infty }$
, (
$b$
)
$\overline {v}/U_{\infty }$
; (
$c$
)
$\overline {u'u'}/U_{\infty }^2$
, (
$d$
)
$\overline {v'v'}/U_{\infty }^2$
, (
$e$
)
$\overline {u'v'}/U_{\infty }^2$
and (
$f$
) variation of
$\overline {u}/U_{\infty }$
and
$\overline {v'v'}/U_{\infty }^2$
along the wake centreline. In (
$a$
–
$e$
), the top half (
$y\geqslant 0$
) is for the flow field associated with
$f_{01}$
, and the bottom half (
$y\lt 0$
) with
$f_{02}$
; for direct comparison. In (
$b$
),
$-\overline {v}/U_{\infty }$
is plotted for
$f_{02}$
mode in the bottom half (
$y\lt 0$
). In (
$a$
,
$b$
), red solid curves mark the recirculation bubble region.

The polygonal cylinder shares similar characteristics of vortex shedding with the circular cylinder in the subcritical flow regime, in terms of the primary tone around
$St \approx 0.2$
for most cases shown in figure 2. However, the geometric modification to polygonal shape strongly affects the flow behaviour including the time-mean width of the wake. Xu et al. (Reference Xu, Zhang, Gan, Li and Zhou2017) and Wang et al. (Reference Wang, Xu, Gan, Zhang and Zhou2019) showed that this altered wake width
$D_w$
can be regarded as an equivalent outer diameter of the cylinder to replace the characteristic length scale for the Strouhal number, yielding
For the two vortex-shedding modes of 7C,
${\textit{St}}_{01}(D) \approx 0.17$
and
${\textit{St}}_{02}(D) \approx 0.24$
, with the corresponding wake widths of
$D_{w,01} \approx 0.98D$
and
$D_{w,02} \approx 0.64D$
, respectively (see figure 13
$c$
). Substituting these values into (4.1) yields
The Strouhal number based on
$D_w$
is now approximately constant, with
${\textit{St}}_{01,02}(D_w) \approx 0.16$
for both tones of 7C, thereby validating
$D_w$
as an effective characteristic length scale, and supporting Xu et al. (Reference Xu, Zhang, Gan, Li and Zhou2017) and Wang et al. (Reference Wang, Xu, Gan, Zhang and Zhou2019). This also suggests that both tones are associated with the same type of vortex-shedding behaviour.
As shown in figure 13(
$d$
), contours of
$\overline {v'v'}(f_{01})$
and
$\overline {v'v'}(f_{02})$
each display symmetric pattern about
$y=0$
, and the peak intensity and pattern distribution are also similar to each other. Their variation along
$y=0$
is quantitatively presented in figure 13(
$f$
), where the statistically based vortex formation length, denoted as
$L_f$
, can be determined at the
$x$
-coordinate for the maximum
$\overline {v'v'}$
, which is also the location of maximum turbulent kinetic energy,
$\text{TKE} = (\overline {u'u'}+\overline {v'v'})/2$
, along
$y=0$
due to the much stronger intensity of
$\overline {v'v'}$
than that of
$\overline {u'u'}$
. Figure 13(
$f$
) thus provides two key insights. That is, firstly, the amplitude of
$\overline {v'v'}$
is similar between the
$f_{01}$
and
$f_{02}$
fields, supported by the contours in figure 13(
$d$
), and secondly,
$L_f(f_{01})\approx 2D$
, which is perceptibly larger than
$L_f(f_{02})\approx 1.76D$
, representing a minor difference of approximately
$12\,\%$
.
Taking into account the first insight, an inspection of figure 13(
$c$
,
$e$
) suggests that stronger
$\overline {u'u'}$
and
$\overline {u'v'}$
values of the
$f_{01}$
field are mainly attributed to the more intense streamwise velocity fluctuation of
$u'(f_{01})$
than of
$u'(f_{02})$
. Additionally, the
$f_{01}$
mode also results in a wider turbulent wake than
$f_{02}$
, with the region of large amplitude further away from the wake centreline. Therefore, it can be inferred from figure 13(
$c$
–
$e$
) that, while the vortex roll-up distances
$L_f$
of the two modes are not very different, the second factor proposed in § 4.2 is validated; namely, mode
$f_{01}$
is more energetic than mode
$f_{02}$
.
Also included in figure 13(
$f$
) is the variation of
$\overline {u}$
along
$y=0$
for both
$f_{01}$
and
$f_{02}$
modes, where the length of the recirculation bubble,
$L_r$
, can be inferred from the saddle point location, i.e.
$\overline {u}(L_r,0)=0$
. The bubble length approximately agrees with
$L_f$
, with
$L_r(f_{01})\gt L_r(f_{02})$
, approximately
$1.65D$
and
$1.43D$
, respectively. Returning to figures 10 and 11(
$c$
), the former shows that, at
$\textit{Re}=4\times 10^4$
, the
$f_{02}$
mode, which has a shorter
$L_r$
, clearly dominates. Thus, the longer
$L_r$
at this Reynolds number, evidenced in the latter figure, is plausibly due to the
${\textit{Re}}$
effect only, rather than directly to a higher probability of mode
$f_{02}$
.
4.4. Distribution of quadrupole sound sources
Following Gloerfelt et al. (Reference Gloerfelt, Pérot, Bailly and Juvé2005), the aerodynamic noise produced by a cylinder can be modelled as two parts, namely, the direct acoustic pressure
$\hat {p}_d$
radiated by quadrupole volume sources in the wake, and the scattered acoustic pressure
$\hat {p}_s$
by the cylinder surface which is of dipole sound nature. By employing a tailored Green’s function
$\hat {G}_1$
that accounts for the presence of the cylinder body, such that
$\partial \hat {G}_1 / \partial n =0$
on the body surface, the direct and scattered acoustic pressure fields can be expressed as follows:
where
$\hat {G}_0$
denotes the free-field Green’s function, and
$\hat {G}_s$
is the component of
$\hat {G}_1 = \hat {G}_0 + \hat {G}_s$
that accounts for the diffraction effects. It is obvious from (4.3) and (4.4) that the quadrupole sources characterised by Lighthill’s stress tensor
$T_{\textit{ij}}$
are the origin of both direct and scattered sound. For high Reynolds number flows as in this study, by neglecting the contribution of viscous force and assuming incompressible and isentropic flow, Lighthill’s stress tensor can be approximated as
$T_{\textit{ij}} \approx \rho _\infty u^{\prime}_i u^{\prime}_j$
.
Under a further assumption of compact acoustic sources (Gloerfelt et al. Reference Gloerfelt, Pérot, Bailly and Juvé2005; Zamponi et al. Reference Zamponi, Avallone, Ragni, Schram and Van Der Zwaag2024), the quadrupole sources distributed in the wake can be represented as a point quadrupole located at the position of maximum TKE, viz., at
$x=L_f$
along
$y=0$
as discussed in § 4.3. The strength of this equivalent quadrupole source,
$\hat {A}$
, is determined by a volume integral of TKE
The amplitudes of the direct and scattered sound fields,
$\hat {p}_d$
and
$\hat {p}_s$
, are thus dependent on
$T_{\textit{ij}}$
and TKE in the cylinder wake, which are characterised by the dominant contribution at the vortex-shedding frequency.
For the dual tones of the 7C cylinder examined in the present study, the noise source regions of the two shedding modes can be modelled as two separate quadrupole sources each located at the respective maximum TKE location, corresponding to
$L_f(f_{01})\approx 2D$
and
$L_f(f_{02})\approx 1.76D$
(see figure 13
$f$
). The assumption of acoustic compactness is satisfied for the source regions since
$kL_f \approx 0.1 \ll 1$
for both shedding modes, where
$k=\omega /c_0$
denotes the wavenumber. Following the frequency segregation method described in § 4.3, the spectrally averaged TKE fields for the two modes
$f_{01}$
and
$f_{02}$
are illustrated in figure 14, along with the overall TKE field produced by both modes. Based on the planar PIV employed in this study, the source strength of the equivalent quadrupole,
$\hat {A}$
, can be estimated through a 2-D approach (Zamponi et al. Reference Zamponi, Avallone, Ragni, Schram and Van Der Zwaag2024), i.e.
Contours of TKE,
$(\overline {u'u'}+\overline {v'v'})/(2U^2_\infty )$
, for 7F and 7C at
$\textit{Re}=2.5\times 10^4$
. Panels show (
$a$
) 7F, (
$b$
) TKE associated with the
$f_{01}$
and
$f_{02}$
modes of 7C, (
$c$
) overall TKE of 7C, (
$d$
,
$e$
) variations of integrated TKE along the
$x$
- and
$y$
-axes, denoted by the notations
$|_y$
and
$|_x$
as calculated by (4.7a) and (4.7b), respectively.

Figure 14(
$b$
) illustrates the normalised TKE distributions, denoted as
$\text{TKE}^* = (\overline {u'u'}+\overline {v'v'})/(2U^2_\infty )$
, for the
$f_{01}$
and
$f_{02}$
modes of 7C, separately. Consistent with the observations in figure 13, the maximum TKE occurs at the position of
$x=L_f$
since the TKE amplitudes are dominated by the transverse component
$\overline {v'v'}$
. The
$f_{01}$
mode shows discernibly higher TKE than
$f_{02}$
due to the stronger intensity of
$\overline {u'u'}$
and similar intensity of
$\overline {v'v'}$
; see figure 13(
$c$
,
$d$
). Comparing the overall TKE fields between 7F and 7C in figures 14(
$a$
) and 14(
$c$
), it is evident that the overall intensity of turbulent fluctuations in the wake is much attenuated for 7C. Figure 14(
$d$
,
$e$
) further presents the variations of integrated TKE along the
$x$
- and
$y$
-axes, respectively, defined as
which confirm that the TKE intensity at
$f_{01}$
is higher than at
$f_{02}$
, and that 7F exhibits a stronger overall TKE than 7C, as well as than both the
$f_{01}$
and
$f_{02}$
modes. Figure 14(
$d$
) also indicates very similar locations of maximum TKE, i.e.
$L_f \approx 1.9D$
for both 7F and 7C, which is an averaged
$L_f$
value between the two shedding modes.
Noise reduction of the 7C case as discussed in § 3.1, compared with the circular cylinder, as well as 7F (since
$\Delta {\textrm{OASPL}}\gt 0$
for 7F; see figure 5), can be explained in terms of the TKE distributions in figure 14 that dominate the equivalent quadrupole source in the wake; see (4.6). According to Gloerfelt et al. (Reference Gloerfelt, Pérot, Bailly and Juvé2005), the quadrupole source strength determines both the direct and scattered acoustic fields (i.e.
$\hat {p}_d$
and
$\hat {p}_s$
), while the source location further affects the diffraction of dipole sound by the cylinder surface. From the TKE flow fields shown in figure 14, the noise reduction mechanism of 7C can be ascribed to the significant attenuation of the quadrupole source strength, for both shedding modes, with nearly no contribution from the source location.
4.5. Proper orthogonal decomposition of each shedding mode
Proper orthogonal decomposition (POD) was first introduced in the context of turbulence by Lumley (Reference Lumley1967), and Sieovich (Reference Sieovich1987) further extended Lumley’s work by developing the method of snapshots POD, which greatly enhanced the computational efficiency of the POD analysis. We now examine the coherent vortex structures and their evolution for modes
$f_{01}$
and
$f_{02}$
, separately, using a snapshot-based POD method and combining both
$u'$
and
$v'$
components, as described in Oudheusden et al. (Reference Oudheusden, Scarano, Hinsberg and Watt2005) and Wang et al. (Reference Wang, Gan, Xu and Zhou2020). Similar to § 4.3, only the PIV snapshots categorised to the shedding mode
$f_{01}$
or
$f_{02}$
are used for the POD analysis of each mode. The POD modes and the distribution of the modal energy content are presented in figure 15. Both shedding modes
$f_{01}$
and
$f_{02}$
exhibit a POD modal energy distribution similar to the circular cylinder (Maryami et al. Reference Maryami, Arcondoulis and Liu2024) and other polygonal cylinders in principal orientations (Wang et al. Reference Wang, Gan, Xu and Zhou2020). That is, the two leading POD modes contain distinctively higher turbulence energy than the subsequent POD modes, which suggests that both
$f_{01}$
and
$f_{02}$
shedding modes are Kármán type. This is further supported by the patterns of the two leading POD modes shown in figure 15(
$b$
–
$e$
), which are evidently spatially translational, in a manner similar to the single primary mode of vortex shedding behind most other polygonal cylinders (Wang et al. Reference Wang, Gan, Xu and Zhou2020).
The POD analysis for the shedding modes
$f_{01}$
and
$f_{02}$
at
$\textit{Re}=2.5\times 10^4$
. (
$a$
) Modal energy distribution according to POD eigenvalues, (
$b$
,
$d$
) the leading POD mode for
$u'$
and
$v'$
, (
$c{,}e$
) the secondary POD mode for
$u'$
and
$v'$
, respectively. In (
$b$
–
$e$
), the top half (
$y\geqslant 0$
) is for shedding mode
$f_{01}$
, and the bottom half (
$y\lt 0$
) for
$f_{02}$
.

Individually, the two shedding modes exhibit noticeable discrepancies. The energy content in the two leading POD modes is slightly higher for the
$f_{01}$
mode than for
$f_{02}$
, as shown in figure 15(
$a$
). In the first 30 POD modes presented, the total energy content for the
$f_{01}$
shedding mode is 80.1 %, higher than that for the
$f_{02}$
mode to a small extent, being 78.6 %. This suggests that vortices shed in mode
$f_{01}$
and mode
$f_{02}$
have similar overall coherency, with the shedding of mode
$f_{01}$
marginally more organised.
Figure 15(
$b$
,
$c$
) illustrates the pattern for the
$u'$
component of the two leading POD modes, with
$f_{01}$
shedding on the top half and
$f_{02}$
shedding on the bottom. The corresponding POD modes for the
$v'$
component are shown in figure 15(
$d$
,
$e$
). Comparison of the two shedding modes reveals a clear difference in terms of the wavelength of the modal distribution, i.e. the streamwise distance between the contour patches of the same colour, which is inversely proportional to
${\textit{St}}_{01,02}$
, within expectation. The size of the patches is visibly larger in the
$f_{01}$
half, implying more energetic vortex shedding of the
$f_{01}$
mode. By calculating the pattern wavelength
$\lambda _m$
for mode
$f_{01}$
and
$f_{02}$
, the ratio
$\lambda _m(f_{02})/\lambda _m(f_{01})\approx 0.68$
, aligned with the
$St$
ratio
${\textit{St}}_{01}/{\textit{St}}_{02}\approx 0.71$
. The vortex formation length can also be inferred from the downstream distance of the first patch in the wake, which suggests
$L_f(f_{01})\gt L_f(f_{02})$
. These observations are in agreement with the findings in § 4.3.
4.6. Evolution of coherent vortices
Due to the relatively high
${\textit{Re}}$
, the instantaneous vortices are distorted and fragmented, POD-based phase averaging is thus employed to exploit the evolution of coherent vortex shedding into the wake, for each of the
$f_{01}$
and
$f_{02}$
modes. Owing to their quasi-periodic nature, phase of each shedding mode is reflected in the coefficients,
$a_1$
and
$a_2$
, associated with the first two POD modes, which obey approximately
subject to turbulence and various levels of vortex structure coherency. Here,
$\lambda _1$
and
$\lambda _2$
are the eigenvalues corresponding to the first two POD modes, and
$\psi _i$
is the phase angle for the
$i$
th snapshot. This gives
where
$r_i$
is the radius in the POD coefficient space for the
$i$
th snapshot.
Phase-average analysis for the shedding pattern of modes
$f_{01}$
and
$f_{02}$
at
$\textit{Re}=2.5\times 10^4$
. (
$a$
) Scatter plot based on the first two POD modes
$a_{1,i}$
and
$a_{2,i}$
. Each dot represents a snapshot. The top half is for mode
$f_{01}$
and the bottom half for
$f_{02}$
. The shaded sector represents the bin size of
$\xi =\pi /8$
, while the red dash-dotted line indicates the phase angle of the corresponding bin. Here,
$a_{1,i}$
and
$a_{2,i}$
are scaled arbitrarily to a mean radius of unity shown by the red circle. (
$b$
–
$i$
) Visulisation of a completed shedding cycle based on
$\lambda _{ci}$
, in
$\pi /2$
phase difference. The middle column with the purple shading 7C is for mode
$f_{01}$
, and the right-most column with the grey shading 7C is for mode
$f_{02}$
with approximately matched phase in each row. In (
$b$
,
$c$
), the dashed circles have equivalent diameter
$D_r$
with their centres marked by the white square, determined following the method in Wang et al. (Reference Wang, Gan, Xu and Zhou2020). The straight dashed lines illustrate the approximate trajectory of vortex cores downstream.

Figure 16(
$a$
) shows the first two POD modal coefficients
$a_1$
and
$a_2$
, with mode
$f_{01}$
in quadrants 1–2, and mode
$f_{02}$
in quadrants 3–4, respectively. Each point represents an instantaneous snapshot of the flow field. The distribution of
$a_1$
and
$a_2$
along the circle of average radius
$r_{\textit{mean}}$
(shown in red) is evident. Snapshots having smaller radii typically exhibit clearer vortex structures, while in those with larger radii, vortices are often distorted and fragmented. As observed in figure 16(
$a$
), snapshots are more scattered in the top half of the POD coefficient space, e.g. towards small radii, which is in line with the finding that mode
$f_{01}$
is more energetic, viz., more turbulent and less coherent compared with mode
$f_{02}$
. The bin size of
$\xi = {\pi }/{8}$
is chosen to calculate the phase average, balancing phase accuracy with the typical number of samples in each bin. This corresponds to approximately
$120$
and
$50$
samples in each bin for shedding modes
$f_{01}$
and
$f_{02}$
, respectively.
Evolution of the coherent vortices in both shedding modes is quantified following a reliable watershed-based vortex identification and segmentation method detailed in Wang et al. (Reference Wang, Gan, Xu and Zhou2020). Vortex centre
$(x_c,y_c)$
is determined by the weighted centroid of the properly isolated and thresholded
$\lambda _{ci}$
contour for each vortex packet, whose equivalent diameter is denoted as
$D_r$
. Here,
$\lambda _{ci}$
is the imaginary part of the complex eigenvalue of the velocity gradient tensor (Zhou et al. Reference Zhou, Adrian, Balachandar and Kendall1999). Circulation
$\varGamma$
of a vortex packet is calculated by integrating
$\omega$
amplitude inside this circle of diameter
$D_r$
, examples of which are marked in figure 16(
$b$
,
$c$
), alongside with the marks for the vortex centres. Using this circle to approximate the vortex domain is primarily for isolating it from the attached shear layer before the former is completely shed. Vortex structure after the phase averaging becomes more discernible. Longer vortex formation length
$L_f$
for mode
$f_{01}$
shedding is reaffirmed, perhaps more conspicuously by inspecting the vortex roll-up distance, comparing (
$b$
) with (
$c$
), or vortex-shedding distance comparing (
$d$
) with (
$e$
). Differences in vortex size
$D_r$
and wavelength, i.e. the distance between adjacent vortices, comparing (
$h$
) with (
$i$
), are also clear.
(
$a$
) Centre trajectories
$y_c$
of the vortices shed from the top surface of 7C at
$\textit{Re}=2.5\times 10^4$
. Trajectories in the bottom half are symmetric. (
$b$
) Variation of the vortex circulation
$\varGamma ^*=\varGamma /(U_\infty D)$
and the equivalent vortex diameter
$D_r$
along the
$x$
-axis, legends follow (
$a$
). In (
$b$
),
$\varGamma _{\textit{max}}/(U_\infty D)\approx 2.25$
and
$1.65$
, respectively, for shedding modes
$f_{01}$
and
$f_{02}$
;
$D_0(f_{01})\approx 1.05D$
and
$D_0(f_{02})\approx 0.84D$
.

Vortex formation length
$L_f$
, which is determined in § 4.3, is commonly agreed to be an important length scale to characterise the wake behind a bluff body. For this reason,
$L_f(f_{01})$
and
$L_f(f_{02})$
of the two shedding modes are used to normalise the
$x$
-coordinate for the quantities plotted in figure 17, viz., vortex centre trajectories
$y_c$
and the evolution of
$\varGamma$
and
$D_r$
. The reasonable data collapse, especially in figure 17(
$b$
) and in the
$x\lesssim L_f$
range in figure 17(
$a$
), evidences the
$L_f$
scaling, which quantitatively lends support to the vortex-shedding visualisation shown in figure 16(
$b$
–
$f$
), that both
$f_{01}$
and
$f_{02}$
shedding modes are indeed similar to the single-mode Kármán type shedding behind most other polygonal cylinders (Wang et al. Reference Wang, Gan, Xu and Zhou2020).
Indeed, for both shedding modes, vortex formation length
$L_f$
seems to divide the behaviour of each examined quantity into two stages. For
$x\lt L_f$
,
$y_c$
rapidly reduces as shown in figure 17(
$a$
), and the centre of the vortex, which is rolling up, moves towards the wake centreline. The rolling-up process is seconded by the expeditious growths of
$\varGamma$
and
$D_r$
in figure 17(
$b$
). At
$x\approx L_f$
,
$\varGamma$
climaxes at
$\varGamma _{\textit{max}}$
, and
$D_r$
at this location is denoted as
$D_0$
. It is found that
$\varGamma _{\textit{max}}(f_{01})/\varGamma _{\textit{max}}(f_{02})\approx 1.36$
, which explains the more energetic
$f_{01}$
shedding mode shown in figure 13, and
$D_0(f_{01})/D_0(f_{02})\approx 1.25$
. The weaker circulation in mode
$f_{02}$
finally leads to reduced far-field noise, as mentioned in § 3.1.1. Since
$\varGamma \sim \omega D^2_r$
, this suggests that the (maximum) phase-averaged vorticity amplitude inside a vortex packet at
$L_f$
is slightly lower for the
$f_{01}$
mode, about
$87\,\%$
of mode
$f_{02}$
.
For
$x\gt L_f$
, however, a clear trajectory split can be seen. Vortices shed by the more energetic
$f_{01}$
mode appear to move away from the centreline very slowly, while vortices of mode
$f_{02}$
have a motion in the opposite
$y$
-direction, but also very mildly. The two modes of shedding show a similar rate of variation in terms of
$\varGamma$
and
$D_r$
until
$x\approx 1.5L_f$
. After that, the two modes start to show weak discrepancies in trends, but both mainly governed by a combined process of vorticity diffusion and dissipation.
4.7. Near-wall flow field
We now identify the origin of the two modes of vortex shedding from the near-field time-resolved PIV measurements, acquired from the small FOV1 in figure 1(
$b$
). Figure 18 presents the near-wall spectrally averaged vorticity field, obtained following the same procedure as described in § 4.3. In shedding mode
$f_{01}$
, figure 18(
$a$
), the shear layer separates at the primary separation point, marked as PSP, without reattachment. While in shedding mode
$f_{02}$
, figure 18(
$b$
), after the primary separation at PSP, the shear layer reattaches to the cylinder surface and then separates at the downstream corner, which is the secondary separation point, denoted as SSP.
Near-wall spectrally averaged vorticity field at
$\textit{Re}=2.5\times 10^4$
for (
$a$
) shedding mode
$f_{01}$
(shear-layer separation), and (
$b$
) shedding mode
$f_{02}$
(shear-layer reattachment). Here, PSP and SSP stand for primary separation point and secondary separation point, respectively.

Apparently, as discussed in § 3.3, the occurrence of mode transition is rather stochastic and unpredictable. A simple periodically reattached shear layer does not result in a dual-tone phenomenon like the one observed in this study, which has been demonstrated in a flow around the square cylinder (Bai & Alam Reference Bai and Alam2018). The stochastic nature of mode
$f_{02}$
is plausibly related to the oscillatory shear-layer flapping motion discussed in Masoudi et al. (Reference Masoudi, Sims-Williams and Gan2023), which is also highly unstable.
Reattachment of the shear layer modifies the vortex roll-up process, which manifests itself in all the quantities discussed above for mode
$f_{02}$
, such as
$L_f$
and
$\varGamma _{\textit{max}}$
. For instance, it reduces the ‘shooting angle’ of the separated shear layer with respect to the wake centreline, which has a direct effect on the vortex trajectory, as shown in figure 17(
$a$
), and in turn, leads to temporary wake contraction. Reduced wake width gives rise to diminished drag coefficient and therefore elevated frequency attributing to their inverse correlation in general (Xu et al. Reference Xu, Zhang, Gan, Li and Zhou2017), which is consistent with the higher
${\textit{St}}_{02}$
than
${\textit{St}}_{01}$
.
4.8. Dipole sound source location
The location of equivalent dipole sound sources on the cylinder surface can be approximately estimated by a cross-correlation analysis of pressure signals between two surface microphones. Unlike spectral averaging for velocity components, acoustic signal processing requires temporal continuity. For this reason, in the time segments based on CWT scalograms of the two modes shown in figure 12(
$e$
), time interval
$\Delta t_1 = 1$
–
$1.4$
s is selected for mode
$f_{01}$
and
$\Delta t_2 = 0.7$
–
$0.9$
s for mode
$f_{02}$
. For the free-stream velocity of
$U_\infty =15\,\textrm {ms}^{-1}$
at
$\textit{Re}=2.5\times 10^4$
, there are at least 30 completed shedding cycles within a 0.2 s sample duration for mode
$f_{02}$
(
$\sim 150$
Hz). The cross-correlation function is expressed as
\begin{equation} R_{si,j}(\tau ) = \frac {p^{\prime}_{si}(t) p^{\prime}_{sj}(t + \tau )}{p^{\prime}_{si, {rms}} \,p^{\prime}_{sj, {rms}}}, \end{equation}
where
$p^{\prime}_{si}$
and
$p^{\prime}_{sj}$
are the pressure fluctuations measured at surface positions
$i$
and
$j$
, respectively, and
$\tau$
denotes the time delay between the two pressure signals.
Cross-correlation
$R_{si,j}$
of surface pressure fluctuations between S2, S4 and S6 with S6 as the reference,
$\textit{Re}=2.5\times 10^4$
. (
$a$
) Shedding mode
$f_{01}$
, (
$b$
) shedding mode
$f_{02}$
.

Figure 19 shows the cross-correlation results for surface pressure fluctuations for both shedding modes
$f_{01}$
and
$f_{02}$
. Surface microphones S2, S4 and S6 are selected for demonstration, with S6 being the reference. It can be observed in figure 19(
$a$
) that, despite some slight shift among the three correlation functions
$R_{s}$
, they share a common period of
$\tau ^*=\tau U_{\infty }/D\approx 5.82$
, corresponding to the frequency of the first tone, i.e.
${\textit{St}}_{01}\approx 0.17$
. In contrast, the period of the three
$R_{s}$
functions in figure 19(
$b$
) reduces to
$\tau ^*\approx 4.0$
, which is equivalent to
${\textit{St}}_{02}\approx 0.24$
of the second tone. These results demonstrate that the acoustic signals were effectively separated into two different spectral frequencies in the cross-correlation analysis.
Furthermore, difference in time delays can be observed between the microphone pairs. In figure 19(
$a$
), the signal in shedding mode
$f_{01}$
is first detected by S4 (
$\tau ^*=-0.32$
), followed by S6 (
$\tau ^*=0$
) and finally by S2 (
$\tau ^*=0.32$
), indicating that the region of equivalent dipole sound source is closest to S4. This order suggests that signal propagates more quickly downstream due to convection by the incoming flow, with S6 detecting the signal sooner than S2. Inoue & Hatakeyama (Reference Inoue and Hatakeyama2002) found that the dipole sound source in the circular cylinder flow is located near the shear-layer separation point, and therefore the sound source of
$f_{01}$
shedding mode is located near PSP (cf. figure 18
$a$
). For shedding mode
$f_{02}$
, figure 19(
$b$
) shows that the order of signal receiver is S6 (
$\tau ^*=0$
), S4 (
$\tau ^*=0.32$
) and S2 (
$\tau ^*=0.49$
), suggesting that the dipole sound source is somewhere near S6 in this case, with the sound wave propagating upstream. The dipole sound source region is now shifted downstream, close to SSP (cf. figure 18
$b$
).
5. Discussion of vortex-shedding mode switching
Finally, a hypothesis is proposed regarding the mechanism of shear-layer separation and reattachment of the two shedding modes, as sketched in figure 20. Small upstream disturbances are inevitable in wind tunnel experiments, despite the very low turbulence intensity level (
$\lt 0.15\,\%$
at
$U_{\infty }=50$
$\textrm {ms}^{-1}$
) of the SUSTech anechoic wind tunnel (Yang et al. Reference Yang, Liu, Liu, Shen, Zhang, Wei, Liu and Xu2021). Generally, flow around the heptagonal cylinder separates at PSP and sheds vortices into the wake, i.e. in a manner of mode I. For low
$\textit{Re}\lt 2\times 10^4$
, the wake width of 7C is large and vortex shedding behaves relatively stably, making the flow rather insensitive to incoming disturbances. This explains the single primary tone observed in the spectra of far-field noise and surface pressure fluctuations in §§ 3.1 and 3.2. The equivalent dipole sound sources, manifested as pressure fluctuations distributed over the cylinder surface which are originated from the scattering of quadrupole sources in the wake, can be regarded as centred on the centreline at the streamwise position of PSP.
Schematic diagram illustrating the switching mechanism of shear-layer separation (mode I) and reattachment (mode II) of the dual-tone phenomenon for 7C.

As
${\textit{Re}}$
increases in the range of
$2\times 10^4$
to
$7\times 10^4$
, the shear layer becomes increasingly unstable. The wake begins to contract in width, and the shear layer shifts closer to the centreline, evolving into a stochastic reattachment to the cylinder surface. The separation point then moves towards SSP, in line with the separation angle range of
$78^{\circ }$
–
$90^{\circ }$
for the circular cylinder (Zdravkovich Reference Zdravkovich1997), ultimately resulting in the shear-layer separation at SSP (i.e. mode II). The relocation of the separation point shifts the dipole source region downstream, centred at the same
$x$
-coordinate of SSP instead. This is accompanied by a higher frequency of vortex shedding due to the contracted wake width, and thus the appearance of a secondary tone in the spectra of far-field noise, surface pressure and wake flow velocity. The stochastic nature of mode II ultimately triggers a switching mechanism between the two quasi-stable vortex-shedding modes, a process similar to the oscillatory shear-layer flapping observed in the LES 7C flow by Masoudi et al. (Reference Masoudi, Sims-Williams and Gan2023).
Another effect of increasing
${\textit{Re}}$
is manifested in the reduced laminar shear-layer length (Anand & Sarkar Reference Anand and Sarkar2015), leading to an early transition to turbulent shear layer which is less influenced by small upstream disturbances. The transition of the shear layer mostly echoes that described by Roshko (Reference Roshko1969) and Zdravkovich (Reference Zdravkovich1997) with the transition onset location migrating upstream as
${\textit{Re}}$
increases. In the subcritical range (
$\textit{Re}=10^4-10^5$
) investigated here, transition indeed first manifests as perturbations in the free shear layer, which subsequently rolls up into discrete vortex packets before evolving into the alternating shedding pattern. The observed mode switching is consistent with the expected dynamics of shear-layer transition in this regime, thereby justifying our hypothesis that the shear-layer transition length affects the development of vortex-shedding mode.
These two
${\textit{Re}}$
effects, i.e. wake contraction and early laminar-to-turbulent transition, compete in the range of
$2\times 10^4 \lesssim Re \lesssim 7\times 10^4$
. At relatively low
$\textit{Re} \lt Re_T$
, the former effect gradually takes dominance and the probability of mode II keeps increasing (see figure 10), which, however, starts to decline when
$\textit{Re} \gt Re_T$
, thus making mode I switch back as the dominant one. When
${\textit{Re}}$
exceeds
$7\times 10^4$
, the laminar shear layer is significantly shortened with a nearly fully developed turbulent shear layer, which is hardly affected by incoming disturbances. As a result, mode II of shear-layer reattachment would rarely be excited, corresponding to the very low probability of the secondary tone in figure 10, and the 7C cylinder flow maintains as in a stable mode I. Therefore, only a single tone will be observed in the near- and far-field spectra.
6. Summary and conclusions
In this study, the far-field noise is systematically measured for polygonal cylinders of
$N\in {\mathbb{Z}}[3,12]$
in principal orientations (either a face or a corner facing the incoming flow normally) for
$\textit{Re}\sim \mathcal{O}(10^4$
–
$10^5)$
. Within this
${\textit{Re}}$
range, a dual-tone transitional phenomenon is observed exclusively for
$N=7, 11$
in the corner orientation, and for
$N=9$
in the face orientation, with no such phenomenon detected for other
$N$
, orientations or for the circular cylinder. This dual tone leads to a clear suppression effect for the total sound level, OASPL, compared with the circular cylinder case.
The formation mechanism of this dual-tone phenomenon is then further investigated for the heptagonal (
$N=7$
) cylinder in corner orientation, denoted as 7C, using surface microphones and time-resolved planar PIV measurements around the cylinder surface and in the wake. This case reflects that the two tonal peaks that are the most clearly separated and the most distinctive in the noise spectrum, at approximately
${\textit{St}}_{01} = 0.17$
and
${\textit{St}}_{02} = 0.24$
, respectively, resulting in the most effective far-field noise reduction.
Continuous wavelet transform analysis reveals that the dual-tone phenomenon in 7C arises from the continuous switching between the two competitive vortex-shedding modes, in a stochastic manner. Statistical analysis of the two modes indicates that the time stamps for each mode approximately follow a symmetric Gaussian distribution, complementing each other. The results from the surface pressure fluctuation and far-field noise measurements suggest that this dual-tone phenomenon is likely to be three-dimensional, as evidenced by the loss of synchronisation reflected in the CWT scalograms at two successive faces in the vicinity of the primary flow separation point.
Coherent vortex structures resulting from the two shedding modes, and their evolution in the wake flow, are resolved by combining CWT with POD based phase-averaging analysis, which confirms that the dual-tone phenomenon arises from the stochastic switching between the two vortex-shedding modes. In particular, the mode corresponding to the higher frequency,
${\textit{St}}_{02}$
, exhibits smaller vortex structures and a reduced turbulence level compared with the mode associated with
${\textit{St}}_{01}$
, albeit sharing a similar shedding pattern. The observed noise reduction of 7C can be ascribed to the attenuated TKE intensity for both shedding modes, which dominates the strength of quadrupole sources in the wake.
Finally, PIV results from the near-wall flow field reassure us on the origin of the two shedding modes. That is, the stochastic reattachment of the separated shear layer gives rise to the switching between the two quasi-stable vortex-shedding modes. This is further validated through surface microphone cross-correlation analysis showing acoustic source relocation during mode transition, and supports the established correlation between dipole sound sources and the separation point of a shear layer (Inoue & Hatakeyama Reference Inoue and Hatakeyama2002).
Due to limitations of the experimental facility, future work will focus on elucidating the shear-layer transitional mechanism on the polygonal cylinders featuring similar dual-tone phenomenon, such as for
$N=9$
and
$11$
, and confirming its consistency to the findings from the present study.
Funding
This research was supported by the National Natural Science Foundation of China (Grant Nos. 12111530102 and 12272163) and the Royal Society International Exchanges (Grant No. IEC/NSFC/201061).
Declaration of interests
The authors report no conflict of interest.











































































































































































































































