This paper revisits several methods used in iterative decoders to improve the performance of the LTE Turbo code for small information blocks. By considering the iterative decoder as a chaotic system, some explanations are given on the unexpected behavior of several post-processing algorithms. A "dither-inspired" algorithm, called blind candidate decoding, is proposed. This algorithm improves error correction by performing several decoding attempts from a set of perturbed input. Combined with the use of the CRC as an outer code, a performance improvement of up to 1 dB in signal to noise ratio can be obtained for a frame error rate of 10 −3. Moreover, a stopping criterion based on the Euclidean distance between the output and the input is proposed to mitigate the critical problem of undetected errors.