We develop a belief-propagation (BP) decoder for the joint decoding of multiple codewords which belong to the same non-binary LDPC code. Decoding is based on soft information in form of joint channel-posterior probabilities of all codeword symbols. We extend the BP algorithm for q-ary LDPC codes such that the FFT-based check node processing is preserved and the complexity remains manageable. This joint decoding is useful in settings in which multiple codewords are transmitted in a non-orthogonal way over the same channel, including multipleaccess with packet collisions, physical-layer network coding and multi-resolution broadcasting. We show in an example that joint decoding can be far superior to separate decoding.
I. INTRODUCTIONWe consider a joint decoder for multiple codewords of a non-binary LDPC code: different messages are encoded with a q-ary LDPC code and transmitted over a multiple-access channel (MAC) in a non-orthogonal way. The received signal hence depends on various codewords and separate decoding is generally suboptimum. This situation occurs e.g. in the following application areas:• Physical-layer network coding. While two-way relaying is the most obvious example [1]-[4], the considered setting also applies to the multi-way relay channel [5] and to compute-and-forward relaying [6]. • Slotted ALOHA and uncoordinated multiple-access schemes which take full advantage of the received signal and side information for collision resolution [7]-[10]. • Multi-resolution broadcast [11], [12], which transmits multiple data streams to users with different channel qualities or receiver capabilities. In this paper, we extend previous work on the joint decoding of binary LDPC [2], [4] or convolutional codes [13] to non-binary LDPC codes and do not restrict our attention to physicallayer network coding but rather focus on the details of the belief propagation (BP) algorithm for efficient joint decoding of multiple messages. Non-binary LDPC codes have shown significant benefits in channels which cannot be decomposed into binary subchannels without incurring a significant loss in achievable rates, e.g. for MIMO and for higher-order modulation [14], [15]. They have also been proposed for space communications since they outperform their binary counterparts in particular for short to moderate blocklength [16]. Since coding in higher-order Galois fields is imperative in many applications of random linear network coding [17], [18], applying the same field order for channel coding facilitates the integration of both approaches.