Efficient decoders for LDPC codes
View Patent ↗An LDPC decoder, applicable to LDPC codes including codes where check nodes within the same group are connected to a common bit node, successively processes groups of check nodes in a particular iteration, including updating bit nodes in that same iteration responsive to messages generated in response to processing a group of check nodes. Within an iteration, the LDPC decoder may also track the number of unresolved parity check equations, and cease iterating or output to an outer block decoder if that number reaches a local minima or standard minimum, falls below a predetermined threshold, or its rate of change falls below a predetermined threshold, indicating a lack of convergence or false convergence condition. The LDPC decoder may also provide a feedback assist to a demodulator. Also, a novel memory configuration may store messages generated by the decoder in the course of check node processing. The decoder may also employ a novel method for generating check node to bit node messages through a prescribed series of pair-wise computations.
1. A method of iteratively decoding LDPC encoded information comprising:
producing, for an iteration of the method, estimated information;
checking whether the estimated information fully resolves a plurality of parity check equations;
if the estimated information fully resolves the plurality of parity check equations, or if the estimated information for the iteration does not fully resolve the plurality of parity check equations, but the number of unresolved parity check equations achieves a minimum. outputting, as decoded information, the estimated information produced in that iteration and then ceasing iterations; and
if the estimated information for the iteration does not fully resolve the plurality of parity check equations, and the number of unresolved parity check equations does not achieve a minimum, performing at least additional iteration.
2. The method of claim 1 wherein the minimum that must be achieved is a local minimum.
3. The method of claim 1 wherein the minimum that must be achieved is a first, local minimum.
4. The method of claim 1 wherein the minimum that must be achieved is a global minimum.
5. The method of claim 1 wherein the estimated information is written to an output memory every iteration, and the step of ceasing iterations when the number of unresolved parity check equations achieves a minimum avoids unnecessary write operations.
6. A method of iteratively decoding LDPC encoded information encoded with an outer block code, comprising:
producing, in an iteration of the method, estimated information;
if the estimated information does not fully resolve a plurality of applicable parity check equations, so that a number of unresolved parity check equations remain, but the number of unresolved parity check equations falls below a predetermined threshold, decoding the estimated information by a decoder of the outer block code; and
if the number of unresolved parity check equations equals or exceeds a predetermined threshold, performing one or more additional iterations.
7. The method of claim 6 further comprising terminating the LDPC iterations if the outer block decoder successfully decodes the estimated information.
8. A method of iteratively decoding LDPC encoded information comprising:
producing, in an iteration of the method, estimated information;
if the estimated information fully resolves a plurality of applicable parity check equations, outputting the estimated information as decoded information;
if the estimated information does not fully resolve a plurality of applicable parity check equations, so that a number of unresolved parity check equations remain, each unresolved parity check equations defining a check node connected to one or more bit nodes, and each bit node having a corresponding bit estimates, and a false convergence condition is detected, indicating the method. had converged to an invalid code word, (1) applying a shrinking algorithm to updated one or more bit estimates corresponding to one or more bit node that is reflected in at least one of the unresolved parity check equations, and (2) resuming decoding iterations based on the updated one or more bit estimates; and
if the estimated information does not fully resolve the plurality of applicable parity check equations, and a false convergence condition is not detected, performing one or more additional iterations of the method.
9. The method of claim 8 further comprising detecting a false convergence condition when a rate of change of the number of unresolved parity check equations in successive iterations falls below a predetermined threshold.