This page is under constant development. Codes can hopefully be downloaded a few months after publication of the related papers. Thanks for your patience & interest.

• Closed-form block Source separation codes maximizing the contrast (2,4), the sum of squares of fourth-order marginal cumulants for real data and mixtures in presence of unknown noise. References: Vail'89, Chamrousse'91, Eurasip Signal Processing'94. This is the so-called CoM2 algorithm.
  • The original 1989 Matlab codes on Independent Component Analysis by contrast maximization, all merged one after the other in a single file codesICA.txt . The file codesICA.txt contains the following:
  • The function tfuni4.m , separating 2 sources from 2 sensor data
  • The code aci.m performing the sweeps: i.e. the code separating N sources from N (or more) sensors, calling the function tfuni4.m that processes the pairs, and putting the results together.
  • The function tfuniV.m identifying a 2-input 2-output mixture from sensor cumulants. The latter allows to evaluate performances for infinite observation lengths.
  • The function aciValid.m performing the sweeps and calling tfuni4V.m. It separates N sources from N (or more) sensors, only based on cumulants of sources (number not limited, so that interferences simulating non Gaussian noises can be included). References: Eurasip Signal Processing'94
  • The function ecar2.m allowing to measure the gap between 2 matrices up to multiplicative scaling and permutation matrices. References: Eurasip Signal Processing'94
• Closed-form block Source separation codes maximizing the contrast (1,4), the sum of fourth-order marginal cumulants, when sources have a kurtosis of given constant sign. References: ICASSP'97, SPAWC'97, ICA'99 This is the so-called CoM1 algorithm.
  • The function tfuni1.m separating a pair of complex sources.
  • The function tfuni1R.m separating a pair of real sources.
  • The function tfuni1V.m identifying a 2-input 2-output mixture (in presence of an unlimited number of non Gaussian unknown interferences) from sensor cumulants. The latter allows to evaluate performances for infinite observation lengths.
  • The code aci1.m separating N complex sources from N (or more) sensors, that calls tfuni1.m.
  • MeX and C codes for separating complex sources, written by C.Guillemot and O.Grellier in June 2000, and implementing the same algorithms as aci1.m and tfuni1.m.
  • Important note: the algorithm tfuni1.m has been improved in 2001, by noting that the optimal 2x2 unitary matrix can be obtained by computing the EVD of a 3x3 symmetric real matrix. The corresponding function is tfuni2.m . Reference: ICISP'01.
• Codes for identifying a complex mixture of 3 sources received on 2 sensors. References: SPIE'98, ICA'99, IEEE Trans SP'04
• Codes on block blind equalization (not optimized, but show feasability and convergence). In particular:
  Reference: Int. Jour. Adapt. Contr. Sig. Proc. (Wiley), SPAWC'03, ICASSP'04
  • KMA: by kurtosis maximization
  • CMA: by constant modulus
  • CPA: by constant power, or polynomial roots matching
• Code for direct estimation of Carrier Residual xx.m . Reference: IEEE Com. Letters
• Code xx.m of the closed-form blind equalizer of a SISO GSM channel. References: SPIE'98, SPAWC'99
• Code xx.m of the closed-form blind equalizer of a MIMO GSM channel. References: SPAWC'99
• Code xx.m of the closed-form semi-blind equalizer of a SISO GSM channel. References: SPAWC'99
• Code xx.m performing quantics decompositions. References: Eurasip Signal Processing'96
• Matlab codes or Scilab codes to compute generic ranks of various tensors. References: I3S report I3S/RR-2006-29-FR, September 4, 2006. Icassp'08 preprint