Iterative data detection: Bounding performance and complexity reduction

Iterative detection (ID) or turbo decoding is a good approximation to the optimal decoder for a concatenated network of Finite State Machines (FSMs) constructed from constituent codes and ISI channel memory. Iterative detection was used successfully for the first time in the decoding of Parallel Concatenated Convolutional Codes (PCCCs or turbo codes). Applications that have been using iterative detection include Serially Concatenated Convolutional Codes (SCCCs), Trellis Coded Modulation (TCM) over interleaved, frequency-selective fading channels, and Serially Concatenated Continuous Phase Modulation (SCCPM).; The Soft Input Soft Output (SISO) module, which is a basic building block in iterative detection, implements an algorithm performing an update of the reliability information of the input and the output symbols of a FSM. For implementation, the complexity should be reduced. Reduced state SISO (RS-SISO) algorithms are used for complexity reduction. When perfect CSI is not available at the receiver, the SISO module should be able to deal with unknown, and possibly time-varying parameters. SISO modules that jointly estimate the parameters and generate soft information will be referred to as Adaptive SISO (A-SISO) modules. In this paper, the RS-SISO algorithm is used for complexity reduction of the A-SISO module, namely RS-A-SISO algorithms. RS-A-SISO modules are applied to SCCPM over frequency-selective fading channels when perfect CSI is not available and trade-offs between performance and complexity are considered. RS-A-SISO algorithms are analyzed using the SNR evolution technique.; We present upper bounds for the maximum-likelihood decoding performance of turbo-like codes for specific interleavers. Previous research developed upper bounds for turbo-like codes using the uniformly interleaved assumption, which bounds the performance averaged over all interleavers. These upper bounds for specific interleavers are based on the simple bound and estimated weight distributions including the exact several smallest distance terms because if either estimated weight distributions on their own or the exact several smallest distance terms only are used, an accurate bound can not be obtained. The approximate weight distribution is verified by statistical reliability measure.