Ambiguity-Free Broadband DOA Estimation Relying on Parameterized Time-Frequency Transform
Wei Wang, Shefeng Yan, Linlin Mao · 2025
Read Ambiguity-Free Broadband DOA Estimation Relying on Parameterized Time-Frequency Transform by Wei Wang, Shefeng Yan, Linlin Mao (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2503-03691.txt.
Introduction
Wei Wang and co-authors present a cleaner way to estimate the direction of broadband sounds in a 2025 paper. Sparse microphone rows can locate sound sources, but wideband signals from moving sources create ambiguity, where several directions fit the data. Low signal strength makes the problem worse.
Their scheme rests on a parameterized time-frequency transform, tuned to sharpen the signal before direction finding. A frequency-difference step then removes the ambiguity that sparse arrays normally suffer. Tests target realistic, non-stationary sounds like speech and machinery. Better direction estimates feed hearing aids, robot ears, and sonar. The method aims to keep accuracy when noise is high and the array must stay small.
Catalog record
| CREATOR | Wei Wang, Shefeng Yan, Linlin Mao |
|---|---|
| YEAR | 2025 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 24 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2503-03691
Browse the full library · About the Tower of Babel
Full text
The opening of the work, from the library's hosted text:
Ambiguity-Free Broadband DOA Estimation Relying on Parameterized Time-Frequency Transform Thanks: This work was supported in part by the National Natural Science Foundation of China under Grant 62192711 and Grant 62371447. (Corresponding author: Shefeng Yan.) Thanks: W. Wang, S. Yan and J. Yang are with the Institute of Acoustics, Chinese Academy of Sciences, Beijing 100190, China, and also with the University of Chinese Academy of Sciences, Beijing 100049, China (e-mail: [email protected]; [email protected]; [email protected]). Thanks: L. Mao is with the Institute of Acoustics, Chinese Academy of Sciences, Beijing 100190, China (e-mail: [email protected]). Thanks: Z. Sui is with the School of Computer Science and Electronics Engi neering, University of Essex, CO4 3SQ Colchester, U.K. (e-mail: [email protected]). Wei Wang, Shefeng Yan, Linlin Mao Abstract An ambiguity-free direction-of-arrival (DOA) estimation scheme is proposed for sparse uniform linear arrays under low signal-to-noise ratios (SNRs) and non-stationary broadband signals. First, for achieving better DOA estimation performance at low SNRs while using non-stationary signals compared to the conventional frequency-difference (FD) paradigms, we propose parameterized time-frequency transform-based FD processing. Then, the unambiguous compressive FD beamforming is conceived to compensate the resolution loss induced by difference operation. Finally, we further derive a coarse-to-fine histogram statistics scheme to alleviate the perturbation in compressive FD beamforming with good DOA estimation accuracy. Simulation results demonstrate the superior performance of our proposed algorithm regarding robustness, resolution, and DOA estimation accuracy. I Introduction Direction-of-arrival (DOA) estimation has been extensively explored in various fields such as wireless communication, radar and sonar systems [ 1 , 2 , 3 , 4 ] , and gained significant attention in the context of sparse arrays [ 5 ] . However, since the sparse array invokes spatial undersampling, its performance suffers a lot due to DOA ambiguity [ 6 , 7 ] . In contrast to the classic solutions that design array structures to mitigate the ambiguity of DOA [ 8 , 9 ] , frequency-difference (FD) manipulation can shift the beamformer to a lower out-of-band frequency with ambiguity-free DOA estimation [ 11 , 10 ] . Then, to further enhance the angular resolution in FD-based conventional beamforming (FD-CBF) [ 12 ] , FD-based multiple signal classification (FD-MUSIC) [ 13 ] , compressive FD beamforming (CFD) [ 14 ] and FD-based atomic norm minimization (FD-ANM) [ 15 ] have been proposed, where the CFD exhibits super resolution but it relies on high signal-to-noise ratios (SNRs) [ 14 , 11 ] . Since most signals in active detection schemes are non-stationary broadband signals [ 16 , 17 ] , the time-frequency analysis methods have been introduced to improve the performance of array signal processing [ 18 ] , such as short-time Fourier transform (STFT) [ 19 ] and Wigner-Ville distribution [ 20 ] . Nonetheless, since the Heisenberg uncertainty principle limits the STFT-based concentration, its DOA estimation performance degrades significantly at low SNRs. Moreover, the cross terms in Wigner-Ville distribution lead to low DOA estimation accuracy when the array data extracts time-frequency samplings from cross terms. Therefore, the MUSIC has been combined with parameterized time-frequency transform (PTFT), which provides higher time-frequency concentration than STFT via an internally defined kernel function without cross terms [ 21 ] . To the best of our knowledge, time-frequency analysis methods have not been exploited for ambiguity-free DOA estimation yet. Thus, we conceive the PTFT-based unambiguous DOA estimation for non-stationary broadband signals. In addition, incoherently averaging DOA spectra is invoked to obtain the final DOA estimates, whereas the incoherent averaging cannot suppress the cross-term effect induced by FD processing in disturbed environments [ 22 ] . Aiming at the above concerns, we conceive a histogram statistics-aided compressive frequency-difference beamforming (HS-CFD) relying on PTFT to strike ambiguity-free DOA estimation for sparse uniform linear arrays. Our contributions are described explicitly below. \bullet We conceive the PTFT-based FD to achieve better signal energy aggregation compared to the conventional FD, and its output SNR is compared with other counterparts in theory and simulation. Then, the PTFT-based CFD is derived. \bullet A coarse-to-fine histogram statistics technique is conceived to replace conventional incoherent averaging, in which grid searching is executed coarsely and finely to alleviate perturbation and preserve DOA estimation accuracy. \bullet Simulation results validate the above analysis results and demonstrate that the proposed algorithm can strike satisfactory robustness, high resolution, and superior accuracy compared to other counterparts. Notations: (\cdot)^{\top} , (\cdot)^{\ast} , |\cdot| , \odot and \lfloor\cdot\rfloor are transpose, complex conjugation, modulus, Hadamard product and rounding down, respectively. \int(\cdot) denotes an indefinite integral. \angle(\cdot) computes the phase angle, e.g., \angle e^{j\theta}=\theta . {j}=\sqrt{-1} is an imaginary unit. \lVert\cdot\rVert_{p} denotes \ell_{p} -norm term. II Problem Formulation II-A Data Model Assuming that an active detection system transmits a non-stationary broadband linear frequency modulation signal s(t) with start frequency f_{\rm L} , end frequency f_{\rm H} , bandwidth B , pulse duration T and overall signal length T_{\mathsf{all}} . Consider K far-field reflected signals impinge upon a M -sensor array with spacing d>c/(2f_{\rm L}) . The frequency-domain data model is written as \boldsymbol{y}({f})=\sum_{k=1}^{K}\boldsymbol{a}(f,\theta_{k})b_{k}S(f)e^{-{j}2\pi f\tau^{k}}+\boldsymbol{n}(f), (1) where \displaystyle\boldsymbol{y}({f})=\left[Y_{1}(f),Y_{2}(f),\cdots,Y_{M}(f)\right]^{\top}, \displaystyle\boldsymbol{a}(f,\theta_{k})=\left[1,e^{-{j}2\pi f\tau_{2}(\theta_{k})},\cdots,e^{-{j}2\pi f\tau_{M}(\theta_{k})}\right]^{\top}, \displaystyle\boldsymbol{n}(f)=\left[N_{1}(f),N_{2}(f),\cdots,N_{M}(f)\right]^{\top}. Y_{m}(f),S(f),N_{m}(f) corresponds to the frequency spectra of the received signal y_{m}(t) , the original signal s(t) , and the noise signal n_{m}(t) on the m th sensor. b_{k} is the complex amplitude, \theta_{k}\in[-90^{\circ},90^{\circ}] , {\tau^{k}} denotes the time delay between the k th target and the first sensor. and \tau_{m}(\theta_{k})=(m-1)d\sin(\theta_{k})/c with c being the propagation speed. \{n_{m}(t)\}^{M}_{m=1} are independent and identically distributed additive Gaussian white noise. II-B CFD in [ 14 ] Processed by FD, the received array data can be recast as \displaystyle{\boldsymbol{z}(f_{1},f_{2})=\boldsymbol{y}({f_{2}})\odot\boldsymbol{y}^{\ast}({f_{1}}),} (2) with f_{2}=f_{1}+\Delta f . Then, the CFD beamforming is derived by \displaystyle{\boldsymbol{x}=\underset{\boldsymbol{x}\in\mathbb{C}^{N_{\rm grid}}}{\arg\min}\,\lVert\boldsymbol{z}(f_{1},f_{2})-\boldsymbol{A}({\Delta f})\boldsymbol{x}\rVert^{2}_{2}+\mu\lVert\boldsymbol{x}\rVert_{1},} (3) where \boldsymbol{x} denotes the solution that draws a DOA spectrum, \boldsymbol{A}({\Delta f})=\left[\boldsymbol{a}({\Delta f},\theta_{1}),\boldsymbol{a}({\Delta f},\theta_{2}),\cdots,\boldsymbol{a}({\Delta f},\theta_{N_{\rm grid}})\right] is a sensing matrix with N_{\rm grid} discretized grids and \mu>0 is a regularization parameter. \boldsymbol{a}(\Delta f,\theta) in \boldsymbol{A}({\Delta f}) is given by \boldsymbol{a}({\Delta f},\theta)=\boldsymbol{a}({f_{2}},\theta)\odot\boldsymbol{a}^{\ast}({f_{1}},\theta) . The resolution of beamforming improves with \Delta f increasing [ 12 ] , but the maximum \Delta f is limited to c/(2d) for ambiguity-free DOA estimation [ 14 ] . Besides, (K^{2}-K) cross terms inevitably produce artifact DOAs [ 13 ] . Also, it is intractable to accumulate the energy of non-stationary signals via fast Fourier transform (FFT). Hence, we propose the PTFT-based HS-CFD algorithm involving the PTFT-based CFD technique and the coarse-to-fine histogram statistics scheme. III Proposed Method III-A PTFT-based CFD Upon exploiting PTFT to s(t) , we can obtain [ 23 ] g_{s}(t,f_{w})=\int_{-\infty}^{\infty}S(f)\Gamma^{\rm R}(f)\Gamma^{\rm S}(f)H_{\sigma}^{\ast}(f-f_{w})e^{{j}2\pi ft}{\rm d}f, (4) where \Gamma^{\rm R}(f)=e^{-{j}2\pi\int\kappa(f){\rm d}f} , \Gamma^{\rm S}(f)=e^{{j}2\pi f\kappa(f_{w})} and \kappa(f) is the kernel function to be defined later. H_{\sigma}(f-f_{w}) denotes the w th analysis window with center frequency f_{\rm L}\leq f_{w}\leq f_{\rm H} , and a window width of \sigma . For the sake of demonstration, a rectangular window function is exploited, yielding {H_{\sigma}(f-f_{w})=}\left\{\begin{aligned} 1,&\ f_{w}-\frac{\sigma}{2}\leq f\leq f_{w}+\frac{\sigma}{2},\\ 0,&\ {\rm otherwise}.\\ \end{aligned}\right. (5) Then, the stationary phase method is utilized to obtain the approximate frequency spectrum [ 24 ] . {S(f)\approx}{\rm Rect}\left(\frac{f-\frac{f_{\rm L}+f_{\rm H}}{2}}{B}\right)\gamma e^{-j\frac{\pi T(f-f_{\rm L})^{2}}{B}}, (6) where \gamma=\sqrt{T/B}e^{j\pi/4} . According to the definition that \Gamma^{\rm R}(f) rotates signals by subtracting \kappa(f) from the phase of S(f) [ 23 ] , \kappa(f) is given by \kappa(f)=-\frac{T(f-f_{\rm L})}{B}+{\rm C}, (7) where