基于小波多分辨分解的HRP算法的快速實現(xiàn)方法
Fast HRP Algorithm Realization Methods Based on the Wavelet Multiresolution Decomposition
-
摘要: MP類分解具有很好地再現(xiàn)信號內(nèi)部正交稀疏結(jié)構(gòu)的能力,但分解普遍存在預(yù)響應(yīng)和局部特性失配等特點,為此S.Jaggi等人提出HRP算法以獲得更為準(zhǔn)確的信號內(nèi)部結(jié)構(gòu),但面臨更為龐大的運算量。本文結(jié)合小波多分辨快速分解算法提出了在小波域?qū)崿F(xiàn)HRP的快速算法,并進(jìn)一步給出了減少運算時間的HRP的并行算法結(jié)構(gòu)。理論和仿真實驗表明,小波方法與HRP的結(jié)合不但可以大大減少HRP的運算量,而且有助于改善小波分析的結(jié)果,是一種很有前途的信號自適應(yīng)分解和特征提取方法。
-
關(guān)鍵詞:
- HRP算法; 多分辨分解; 稀疏; 局部特性
Abstract: The Matching Pursuit(MP) algorithms display good performance of recurring the orthonormal sparse structure of signals, but the signal decomposition process widely ex-hibits pre-echo artifact and local mismatch, so HRP algorithm was proposed by S. Jaggi, et al. to acquire more exact inner structure of signals. Unfortunely HRP algorithm is followed by more huge operation cost. The fast HRP algorithm is advanced at wavelet domain by tak-ing advantage of wavelet multiresolution decomposition, and a parallel algorithm framework is used to further reduce operation time. Theory and simulation trials indicate that HRP algorithm at wavelet domain not only reduces HRP operation cost greatly, but also improves the effect of the wavelet analysis, thus it is a promising technique applied in adaptive signal decomposition and feature extraction. -
Mallat S, Zhang Z. Matching pursuit with time-frequency dictionaries[J].IEEE Trans. on SP.1993,41(12):3397-3415[2]Jaggi S, Willam C K, Mallat S. High resolution pursuit for feature extraction. Tcch. Rep.LIDS-P-2371, MIT, Nov. 1996.[3]Gribonval R, Bacry E, Mallat S, et al.. Analysis of sound signals with high resolution matching pursuit. in Proc. IEEE-SP Int. Symp. Time-Freq. Time-Scale Anal., Paris, France, June 1996:125-128.[4]Montufar-Chaveznava R, Quantized high resolution pursuit. in Proc. IEEE-SP Int. Symp.Time-Freq. Time-Scale Anal., Pittsburgh, Pennsylvania, U.S.A., 1998: 189-192.[5]楊福生.小波變換的工程分析與應(yīng)用.北京:科學(xué)出版社,2000:148-152. -
計量
- 文章訪問數(shù): 2207
- HTML全文瀏覽量: 100
- PDF下載量: 564
- 被引次數(shù): 0