Benedikt Loesch
Institute for Signal Processing and System Theory, University of Stuttgart
benedikt.loesch@iss.uni.stuttgart.de

NOTE: Runtimes are for unoptimized MATLAB code, so significant speedups could 
be expected for an optimized C implementation

********* Dataset1 ***********

Steps in separation:
- Segmentation using the algorithm from [1].
- Then 4 different algorithms:

Alg 1: TF sparseness based blind beamforming [3,4]

Alg 2: TF sparseness based blind beamforming [3,4] with TF masking post processing

Alg 3: 4x4 FDICA (NFFT=2048) with multidimensional SCT for permutation correction [2]

Alg 4: 4x4 FDICA (NFFT=2048) with multidimensional SCT for permutation correction [2] and some Wiener-Filter like post-processing


Runtime per excerpt on a single core of Intel(R) Core(TM)2 Quad CPU @ 2.40GHz:\

Alg 1. and Alg 2.: MATLAB (50 iterations): 20 minutes, C (50 iterations): ~6 minutes, 
(reducing the number of iterations would speed up)

Alg 3. and Alg 4.: MATLAB: 8 minutes



********* Dataset2 ************


------- 1moving_1fixed -------
Alg 1. Online TF sparseness based blind beamforming [3,4]
Alg 2. Online TF sparseness based blind beamforming [3,4] with TF post processing processing

Runtime per excerpt on a single core of Intel(R) Core(TM)2 Quad CPU @ 2.40GHz:
Alg 1. and Alg 2.: 30 seconds

-------- 1moving_1random ---------
- Segmentation using the algorithm from [1].
- Then 2 different algorithms:

Alg 3. FDICA (NFFT=2048) with multidimensional SCT for permutation correction [2]
Alg 4. FDICA (NFFT=2048) with multidimensional SCT for permutation correction [2] and some Wiener-Filter like post-processing

Runtime per excerpt on a single core of Intel(R) Core(TM)2 Quad CPU @ 2.40GHz:
Alg 3. and Alg 4.: MATLAB (50 iterations): 7 minutes


******** References ************
[1] B. Loesch, and B. Yang: "Adaptive Segmentation and Separation of 
Determined Convolutive Mixtures under Dynamic Conditions", Proc. LVA/ICA 2010
[2] B. Loesch, F. Nesta, and B. Yang: "On the Robustness of the 
Multidimensional State Coherence Transform for Solving the Permutation Problem 
of Frequency-Domain ICA", Proc. IEEE ICASSP 2010, Dallas, USA, March 2010"
[3] B. Loesch, and B. Yang: "Blind Source Separation based on Time-Frequency 
Sparseness in the Presence of Spatial Aliasing", Proc. LVA/ICA 2010
[4] B. Loesch, and B. Yang: "Online Blind Source Separation based on Time-
Frequency Sparseness", Proc. IEEE ICASSP 2009, Taipeh, Taiwan, April 2009