AbstractCoded modulation was first proposed by Ungerboeck in 1982 in which large coding gains can be obtained without compromising bandwidth efficiency. It can be classified into two classes according to the coding strategies: One is Trellis Coded Modulation (TCM) in which convolutional codes are combined with modulation. The other is Block Coded Modulation (BCM) in which block codes are employed. Codes constructed in this way are called BCM codes. The thesis mainly deals with coding and decoding of BCM codes. Main contributions of the work are as follows:I.In the construction of BCM codes, we have overcome the restriction in selecting binary component codes by introducing a new construction method which allows the use of variable block length component codes.Simulation results show BCM codes constructed by this method have better performance and complexity than conventional ones.2.Bit Error Rate (BER) of BCM codes are analyzed over Rayleigh-fading channels. With the results that large path multiplicity, which is caused by Ungerboeck set partitioning method, is a major factor which degrades the code performance significantly. A new set partitioning method that greatly reduces the path multiplicity is presented to overcome the degradation. Simulation of 8-PSK BCM codes with the new set partitioning shows that 0.5 .2dB coding gains can be achieved compared to the conventional codes of the same BER.3.From the point of decoding, a new multistage decoding method is given on the basis of the traditional one. Simulation results show this new method achieves a better bit error performance than conventional multistage decoding with less sacrifice of the decoding complexity.4.Three methods, that is, interleaving between the coded bits of every level, use of reliability information about decoding, and iterative decoding are added to the traditional multistage decoding method to better the performance of BCM code. We show that such a scheme can gain about 1.7dB in signal to noise ratio compared to the traditional multistage decoding at the BER of I o~. Also, when the reliability information is extracted by soft output Viterbi algorithm (SOyA). It has only 50% more complexity than the original Viterbi algorithm and approximately only 25% complexity that of the MAP algorithm.