阮宗利 李立萍 錢國兵 羅明剛
?
基于含噪復值ICA信號模型的快速不動點算法
阮宗利*①②李立萍①錢國兵①羅明剛①
①(電子科技大學電子工程學院 成都 611731)②(中國石油大學(華東)理學院 青島 266580)
復數快速不動點算法亦稱為復數FastICA算法,是盲信號分離的一類重要算法。然而,該算法對被噪聲污染的混合源的分離效果較差,尤其是在低信噪比的情況下。這主要是由于在噪聲環境下,被白化過后的信號樣本的相關矩陣不再是單位陣而是一個對角矩陣。該文基于復信號快速不動點算法,首先將基于含噪復值ICA信號模型的混合源投影到信號子空間,以便進行去噪和去相關處理,然后對現有的復數FastICA算法的學習規則做了修正,從而在迭代更新過程中考慮了噪聲的影響,因此將顯著提高復數FastICA算法的盲信號分離性能。文中給出了去噪非圓信號nc-FastICA算法的推導和步驟,仿真結果說明了該算法的有效性。
獨立分量分析;復數快速不動點算法;圓信號;非圓信號;去噪


然而,目前并沒有能夠較好處理噪聲ICA模型的復數快速不動點算法。張和發等人[15]在2011年對nc-FastICA算法做了修正,并將之用于提取微弱信號,然而其主要是在算法過程的白化步驟中進行了去噪處理,效果不佳。本文提出了去噪復數FastICA算法,對噪聲復數ICA模型的FastICA算法的學習規則做了改進,起到了較好的去噪作用,使得分離效果有顯著提高。
經典的復數ICA信號模型為

一般地,快速不動點算法首先要對觀測數據進行白化處理,即





非圓復信號的快速不動點算法的學習規則為


具有噪聲的復數ICA信號模型如下:




此時,偽白化后的觀測數據應表示為:





從而有了式(7)中右端的第1項。


因此,考慮到噪聲后,對應式(15)有

從而,對應式(7)有

根據前面的分析,可以得到去噪nc-FastICA算法的具體步驟:





圖1 兩個8QAM信號及其混合與分離的散點圖

圖2 兩個CGGD信號的混合與分離

圖3 在不同信噪比下nc-FastICA算法和去噪nc-FastICA算法的性能對比
關于兩個非線性函數效果的差異問題,非線性函數不同,分離效果會有差異,因為分離效果受很多因素的影響,比如信源超高斯、次高斯,非圓系數,收斂條件的設定等。
本文對噪聲復ICA信號模型的快速不動點算法進行了研究。首先通過對觀測信號的協方差矩陣進行信號子空間和噪聲子空間分解,從而在偽白化過程中將觀測信號投影到信號子空間;然后在已有復數FastICA算法的學習規則的基礎上,考慮了噪聲影響,對學習規則做了改進,使之更符合帶噪聲的信號模型。因此,去噪nc-FastICA算法提高了盲辨識性能。仿真結果驗證了本文算法的有效性。
[1] Comon P. Independent component analysis, a new concept? [J]., 1994, 36(3): 287-314.
[2] 陳曉軍, 成昊, 唐斌. 基于ICA 的雷達信號欠定盲分離算法[J]. 電子與信息學報, 2010, 32(4): 919-924. Chen Xiao-jun, Cheng Hao, and Tang Bin. Underdetermined blind radar signal separation based on ICA[J].&, 2010, 32(4): 919-924.
[3] Ranganathan R, Qiu R, Hou Shu-jie,.. Blind recovery of smart meter wireless transmissions employing independent component analysis in the presence of strong wideband interference[C]. IEEE International Conference on Smart Grid Communications (SmartGridComm), Brussels, 2011: 172-177.
[4] 付衛紅, 楊小牛, 劉乃安. 基于四階累積量的穩健的通信信號盲分離算法[J]. 電子與信息學報, 2008, 30(8): 1853-1856. Fu Wei-hong, Yang Xiao-niu, and Liu Nai-an. Robust algorithm for communication signal blind separation fourth-order-cumulant-based[J].&, 2008, 30(8): 1853-1856.
[5] Sun Zhan-li and Lam Kin-man. Depth estimation of face images based on the constrained ICA model[J]., 2011, 6(2): 360-370.
[6] Ma Xin-yue, Zhang Hang, Zhao Xiao-jie,.. Semi-blind independent component analysis of fMRI based on real-time fMRI system[J]., 2013, 21(3): 416-426.
[7] Nafta A, Johannisson P, and Shtaif M. Blind equalization in optical communications using independent component analysis[J]., 2013, 31(12): 2043-2049.
[8] Hyv?rinen A and Oja E. A fast fixed-point algorithm for independent component analysis[J]., 1997, 9(7): 1483-1492.
[9] Bingham E and Hyv?rinen A. A fast fixed-point algorithm for independent component analysis of complex valued signals[J]., 2000, 10(1): 1-8.
[10] Yang Zhang and Kassam S A. Optimum nonlinearity and approximation in complex FastICA[C].46th Annual Conference on Information Sciences and Systems (CISS), Princeton, NJ, 2012: 1-6.
[11] Novey M and Adali T. On extending the complex FastICA algorithm to noncircular sources[J]., 2008, 56(5): 2148-2154.
[12] Li Hua-liang, Correa N M, Rodriguez P A,.. Application of independent component analysis with adaptive density model to complex-valued fMRI data[J]., 2011, 58(10): 2794-2803.
[13] Jouny I. Target recognition using UWB radar signatures extracted via complex ICA[C]. IEEE International Conference on Ultra-Wideband (ICUWB), Syracuse, NY, 2012: 477-481.
[14] Mehrabian H, Chopra R, and Martel A L. Calculation of intravascular signal in dynamic contrast enhanced-MRI using adaptive complex independent component analysis[J]., 2013, 32(4): 699-710.
[15] 張和發, 李立萍, 楊小牛, 等. 一種適用于微弱信號盲提取的白化方法[J]. 電子學報, 2011, 39(6): 1297-1301. Zhang He-fa, Li Li-ping, Yang Xiao-niu,.. An efficient whitening method for weak signal extraction[J]., 2011, 39(6): 1297-1301.
[16] Picinbono B. On circularity[J]., 1994, 42(12): 3473-3482.
[17] Loesch B and Yang Bin. Cramér-Rao bound for circular and noncircular complex independent component analysis[J]., 2013, 61(2): 365-379.
[18] Tong L, Soon V C, Huang Y F,.. AMUSE: a new blind identification algorithm[C]. IEEE International Symposium on Circuits and Systems, New Orleans, LA, 1990, 3: 1784-1787.
[19] Xiao Man-lin, Wei Ping, and Tai Heng-ming. Estimation of the number of sources based on hypothesis testing[J]., 2012, 14(5): 481-486.
[20] Amari S, Cichocki A, and Yang H H. A new learning algorithm for blind signal separation[C]. Advances in Neural Information Processing Systems (NIPS), Colorado, 1995: 757-763.

阮宗利: 男,1978年生,講師,博士生,研究方向為盲源分離、陣列信號處理.
李立萍: 女,1963年生,教授,博士生導師,主要研究方向為陣列信號處理、高速信號處理、微弱信號檢測與參數估計等.
錢國兵: 男,1987年生,博士生,研究方向為盲源分離、陣列信號處理.
羅明剛: 男,1977年生,博士生,研究方向為非合作信號處理、陣列信號處理.
Fast Fixed-point Algorithm Based on Complex ICA Signal Model with Noise
Ruan Zong-li①②Li Li-ping①Qian Guo-bing①Luo Ming-gang①
①(,,611731)②(,,266580)
The complex fast fixed-point algorithm, also called complex FastICA, is one of the most important algorithms for Blind Signal Separation (BSS). However, the performance of this algorithm deteriorates when it is used to separate the noisy mixed sources, especially in the low SNR case, since the covariance matrix of whitened observations is not an identity matrix but a diagonal matrix. This paper bases on the present complex FastICA. First, the mixed sources defined with complex Independent Component Analysis (ICA) signal model are projected onto the signal subspace. Thus, the denoising and decorrelating from mixed signal samples can be handily achieved. Then, the learning rule of the algorithm is modified, where the effect of white Gaussian noise is taken into account. Therefore,the BSS performance of complex FastICA is improved markedly. In this paper, the learning rule of denoised noncircular FastICA (nc-FastICA) is derivated and the detailed procedure is given. Simulation results demonstrate the effectiveness of the proposed algorithm.
Independent Component Analysis (ICA); Complex fast fixed-point algorithm; Circular signal; Noncircular signal; Denoise
TN911.7
A
1009-5896(2014)05-1094-06
10.3724/SP.J.1146.2013.00951
阮宗利 RuanZL0496@sina.com
2013-07-01收到,2013-10-18改回