Gsm Channel Estimator Using A Fixed-Point Matrix Inversion Algorithm

This paper presents a fixed-point matrix inversion implementation of a linear minimum mean square error (LMMSE) channel estimator for GSM receivers. The matrix inversion algorithm uses Cholesky decomposition implemented by a reconfigurable processing element. The fixed-point word length analysis is based on the bit error rate (BER) simulations using standard GSM channel models: TU, HT, and RA. The obtained results are compared to floating point matrix inversion results. Our simulations show that 16 bits fixed-point implementation of matrix inversion in LMMSE estimator gives equal BER performance compared to floating point matrix inversion. The condition number of involved matrices in LMMSE estimator varies depending on the channel model. This variation affects the word length requirements of the matrix inversion engine.