Blind GSM channel estimation under channel coding conditions

Iterative blind and non-blind channel estimation in GSM receivers

“…1 and 5 and our modifications to the full rate data transmission mode TCH/F9.6 described in section III-A, a block of ¾¿¾ (¾ ) bits 4 is channel encoded and interleaved into bursts of ½½ (½ ¾) data bits ´ µ, where the numbers in front of and within the brackets refer to the non-blind and blind case, respectively. Assembled together with the midamble 5 Ñ´ µ, the tail bits, and the guard time, a burst of ½ bits is differentially encoded, modulated, and then propagated through each slice ´ Ø µ of the respective sample channel 6 derived from the stochastic GAUSSIAN STATIONARY UNCORRELATED SCATTERING (GSUS) model [4]. We use standard COST-207 Bad Urban (BU) profiles with maximum Doppler frequencies of Ñ Ü ¼ , ¾¼¼, and ¼¼ ÀÞ, where we take the channel's time-variance within each burst period into account.…”

“…Besides the non-blind LS scheme, we have selected the HOS-based EIGENVECTOR APPROACH TO BLIND IDEN-TIFICATION (EVI) by Boss et al [1] which is a non-linear method maximizing a fourth order cross-cumulant on a second order boundary condition. An investigation of further algorithms based on HOS and SOS can be found in [2,6,7]. A common drawback of all existing blind approaches whether they are based on HOS or SOS is the interference of the estimated channel impulse response by an unknown complex factor.…”

“…A common drawback of all existing blind approaches whether they are based on HOS or SOS is the interference of the estimated channel impulse response by an unknown complex factor. In [2,6] we have presented the so-called BLIND CHANNEL ESTIMATION CORRECTION (BECO) scheme which overcomes this fundamental problem. Based on the a-priori knowledge of the transmitted data symbols ´ µ ¾ ½ ½ , the complex factor can be calculated after filtering the received sequence Ý´ µ with a Mean Square Error (MSE) equalizer followed by time averaging over each burst.…”

“…However, their repeated transmission leaves the communication system with a considerable overhead which could be used for the transmission of additional data sequences, if the channel estimation was solved blindly. In recent publications [2,6], we have shown that some blind approaches yield promising results in estimating GSM channels although there is still a significant AE ¼ loss compared to non-blind channel estimation schemes. In order to reduce this loss, we have utilized an iterative channel estimation which employs the bits after channel decoding as a new "training" sequence.…”