趙寶利



【摘 要】信號(hào)處理中波達(dá)方向估計(jì)在雷達(dá)、聲吶和移動(dòng)通信等領(lǐng)域中具有廣泛的應(yīng)用,其中最具代表性的高分辨子空間算法都是在基于信號(hào)源數(shù)量已知的情況下進(jìn)行的,因此許多目標(biāo)源數(shù)檢測(cè)的算法不斷被提出。以均勻線陣為例,在分析陣列接收數(shù)據(jù)協(xié)方差矩陣特征值的基礎(chǔ)上,通過引入無監(jiān)督學(xué)習(xí)中的聚類算法,完成對(duì)協(xié)方差矩陣特征值的分離,從而有效地檢測(cè)信號(hào)源數(shù)。仿真結(jié)果表明,本文算法在較低的信噪比和較小的快拍數(shù)下相對(duì)傳統(tǒng)算法具有很高的檢測(cè)精度。
【關(guān)鍵詞】波達(dá)方向;無監(jiān)督學(xué)習(xí);信噪比
Research on Detection Algorithm of Number of Signal Sources
ZHAO Baoli
[Abstract] The direction of arrival (DOA) estimation in signal processing is widely used in radar, sonar and mobile communication. The most representative high-resolution subspace algorithm is based on the known number of the signal source. Thus, many target source number detection algorithms have been constantly proposed. In this paper, we take the uniform linear array as the example. Based on the analysis of the eigenvalues of array data covariance matrix, the eigenvalues of covariance matrix will be separated by introducing the clustering algorithm contained in unsupervised learning. Finally, the number of signal sources is effectively detected. The simulation results show that compared to traditional algorithms, our algorithm has a high detection accuracy in the condition of low signal-to-noise ratio (SNR) and small number of snapshots.
[Key words]DOA; unsupervised learning; signal to noise ratio
1 引言
DOA估計(jì)(Direction of Arrival Estimation)[1-5]是空間譜估計(jì)[6-9]以及雷達(dá)信號(hào)處理[10-12]中的關(guān)鍵技術(shù)。研究人員在對(duì)DOA估計(jì)的研究中提出了許多算法,最具代表性的高分辨子空間算法,如MUSIC(Multiple Signal Classification)[13]算法和ESPRIT(Estimation Signal Parameter via Rotational Invariance Techniques)[14]算法一直備受青睞,很多與之相關(guān)的改進(jìn)算法不斷被研究。但這些算法都是在信號(hào)源數(shù)已知的前提下進(jìn)行的[11],在沒有準(zhǔn)確對(duì)信號(hào)源數(shù)目檢測(cè)的前提下,這些算法大多會(huì)失效。而在已知信號(hào)源數(shù)目的情況下大多數(shù)的超分辨DOA估計(jì)算法都具有很好的性能[15]。
在對(duì)信號(hào)源檢測(cè)時(shí),基于統(tǒng)計(jì)信息的信號(hào)源數(shù)目估計(jì)的MDL(Minimum Description Length)準(zhǔn)則[16]、AIC(Akaike Information Theoretic Criteria)準(zhǔn)則[17]被提出,將信息論思想引入到信號(hào)源數(shù)目估計(jì)來降低主觀判斷的影響。其中,AIC準(zhǔn)則不是一致性估計(jì),在小樣本和低信噪比情況下具有過估計(jì)問題;MDL準(zhǔn)則雖然滿足一致性估計(jì),但仍存在欠估計(jì)現(xiàn)象[18]。針對(duì)有色噪聲條件下信號(hào)源數(shù)目估計(jì),文獻(xiàn)[19]與文獻(xiàn)[20]基于蓋氏圓定理提出了蓋氏圓方法。……