Adaptive Channel Coding for Mobile Channels - Or Why Wasting Bandwidth for Error Detection?
Christian H. Weiß, Thomas Stockhammer, Andreas Donner, Joachim Hagenauer · elib (German Aerospace Center) · 2001
We present a new channel coding system which combines error detection and error correction using a single high-memory convolutional code, i.e., no dedicated error detection code is required. Since the state complexity of high-memory convolutional codes makes maximum-likelihood decoding impractical, a modified sequential decoder –the Far End Error Decoder (FEED) – is proposed. FEED adds to the conventional sequential decoding procedure a post processing operation which allows to compute the path reliability of the decoded path. With this path reliability the first error in the decoded frame can be localized, thus, enabling combined error correction and detection with a single code. The advantage of the proposed scheme over applying a dedicated error detection code is demonstrated at the example of the GSM Fullrate Speech Codec. Although the Fullrate Speech Codec is by far not the most suited source coding scheme for our new approach, a gain of up to ¢ dB in SegSNR or £ dB in channel SNR can be achieved with a peak complexity equal to the standard GSM system. However, in average FEED uses a significantly lower number of operations. Moreover, FEED has a scalable complexity and is self-adaptive to varying channel conditions. 1