Bit _ inte . rleaved coded modulations for multiple - input mult iple - output channels

We describe a soft feedback i terative decoding technique for bit interleaved coded modulations (BICM) where the modulator uses multiple t ransmi t t ing antennas and the receiver employs space-time coherent detection wi th multiple antennas. We investigate the bit error rate and the f rame error rate per formance of this technique applied to two families of error-control codes: simple nonrecursive binary convolutional codes and parallel turbo codes based on two recursive constituents. Two different types of fading channels are considered: independent flat Rayleigh and block fading channels. A 6 4 s t a t e rate 1/2 binary convolutional code performs 1.5 dB from outage capacity wi th perfect channel state information and 3 dB wi th Expectation-Maximization estimation on the block fading channel. Iterative a posteriori probability (APP) decoding of compound codes is a very popular and an extremely efficient tool especially when a reasonable complexity maximum likelihood decoder is not available [1][2][3]. Iterative detection is not to be limited to compound error-correcting codes, since it can be applied to any concatenation whose elements are simple enough to make an easy evaluation of the a posteriori probability via the Forward-Backward algorithm [4] or more generally a soft-input soft-output (SISO) decoder [5]. This is the case for the structure of a bit interleaved coded modulation [6] [7] which can be considered as the serial concatenation of an outer error-correcting code and a channel modulator. Multiple-input multiple-output (MIMO) channels permit a wide increase in the systems' spectral efficiency [S][9] and hence to reach modern technologies needs, as unveiled by Tarokh et al. for trellis codes [lo] or more recently for turbo codes [ll]. Combined with bit interleaved coded modulation and APP decoding, this technique is even more efficient. As shown later in this paper, using a 64-state convolutional code of rate 1/2 and a total spectral efficiency of 2 bits per symbol period, the obtained performance is only 1.5 dB away from outage capacity on a block fading channel. The assumption we made according to which the channel state information is known at the receiver end is validated by our implementation of an Expectation-Maximization (EM) [12][13] channel estimator naturally included in the iterative decoding scheme. It is indeed shown that a loss of only 1.5 dB occurs when the receiver recovers information on the channel state on its own via EM.