Using the principles of generalized concatenation we introduce a new design
for coded M-ary CPFSK systems with modulation index h = 1/M. This techniqu
e is based on a set partitioning method of inner CPFSK into nested subsyste
ms with increasing free Euclidean distance which are combined with several
outer convolutional codes having different error-correcting capabilities. I
n other words, we construct multilevel codes on the basis of inner modulati
on with memory, namely CPFSK. The new generalized concatenated construction
s were simulated in an additive white Gaussian noise channel. The multistep
decoding algoritm is described where for soft output decoding the well-kno
wn symbol-by-symbol maximum a posteriori probability (MAP) algorithm is app
lied to the inner CPFSK. A comparison of the generalized concatenated codes
with the best known trellis-coded CPFSK schemes is also presented.