High throughput decoder architecture for low-density parity-check convolutional codes
A Low-Density Parity-Check Convolutional Code (LPDCCC) decoder ( 10 ) for partial parallel decoding of low-density parity-check convolutional codes, the decoder having: a plurality of pipeline processors ( 11 ) to receive channel messages and edge-messages; each processor ( 11 ) having: a plurality of block processing units (BPUs) ( 13 ), each BPU ( 13 ) having a plurality of check node processors (CNPs) ( 14 ) to process check nodes that enter into the processor ( 11 ) and a plurality of variable node processors (VNPs) ( 15 ) to process variable nodes that are about to leave the processor ( 11 ); and a plurality of Random Access Memory (RAM) blocks ( 30 ) for dynamic message storage of the channel messages and the edge-messages; wherein in each processor ( 11 ), the VNPs ( 15 ) are directly connected to corresponding RAM blocks ( 30 ), and the CNPs ( 14 ) are directly connected to corresponding RAM blocks ( 30 ) such that the connections from the VNPs ( 15 ) and CNPs ( 14 ) to the corresponding RAM blocks ( 30 ) are pre-defined and fixed according to a parity-check matrix of an unterminated time-varying periodic LDPCCC.
1. A Low-Density Parity-Check Convolutional Code (LPDCCC) decoder for partial parallel decoding of low-density parity-check convolutional codes, the decoder comprising:
a plurality of pipeline processors to receive channel messages and edge-messages; each processor having:
a plurality of block processing units (BPUs), each BPU having a plurality of check node processors (CNPs) to process check nodes that enter into the processor and a plurality of variable node processors (VNPs) to process variable nodes that are about to leave the processor; and
a plurality of Random Access Memory (RAM) blocks for dynamic message storage of the channel messages and the edge-messages;
wherein in each processor, the VNPs are directly connected to corresponding RAM blocks, and the CNPs are directly connected to corresponding RAM blocks such that the connections from the VNPs and CNPs to the corresponding RAM blocks are pre-defined and fixed according to a parity-check matrix of an unterminated time-varying periodic LDPCCC.
2. The decoder according to claim 1 , wherein RAM blocks of different processors are combined into a plurality larger RAM blocks such that each larger RAM block has a RAM block from each processor.
3. The decoder according to claim 2 , wherein each larger RAM block has an address controller.
4. The decoder according to claim 3 , wherein at least two larger RAM blocks share the same address controller.
5. The decoder according to claim 3 , wherein the address controller is a counter that is incremented by one after every stage of decoding the LPDCCC.
6. The decoder according to claim 1 , wherein the CNP is a look-up table tree.
7. The decoder according to claim 1 , wherein the messages processed by the CNP and from the VNP are updated in the same stage using a shifting operation such that updated messages from the VNP and the channel messages associated with the updated variable nodes are directly output to the next processor.
8. The decoder according to claim 7 , wherein a portion of updated messages from the CNP are not processed by the VNP and are written to local memories.
9. The decoder according to claim 1 , wherein the edge-message is a check-to-variable message or variable-to-check message.
10. The decoder according to claim 1 , wherein the messages are quantized in a few bits to reduce complexity.
11. The decoder according to claim 1 , wherein the VNP is an adding operation using an adder tree.
12. The decoder according to claim 1 , wherein the decoder is implemented on a field-programmable gate array (FPGA) or an Application-Specific Integrated Circuit (ASIC).
13. A method for partial parallel decoding of Low-Density Parity-Check Convolutional Codes (LPDCCCs), the method comprising:
receiving channel messages and edge-messages by a plurality of pipeline processors, and each processor having: a plurality of block processing units (BPUs),
processing check nodes that enter into the processor by check node processors (CNPs) in each BPU;
process variable nodes that are about to leave each processor by variable node processors (VNPs) in each BPU;
dynamically storing the channel messages and the edge-messages in a plurality of Random Access Memory (RAM) blocks in each processor;
directly connecting the VNPs to corresponding RAM blocks in each processor;
directly connecting the CNPs to corresponding RAM blocks in each processor;
wherein the connections from the VNPs and CNPs to the corresponding RAM blocks are pre-defined and fixed according to a parity-check matrix of an unterminated time-varying periodic LDPCCC.
14. The method according to claim 13 , further comprising combining RAM blocks of different processors into a plurality larger RAM blocks such that each larger RAM block has a RAM block from each processor.
15. The method according to claim 14 , wherein each larger RAM block has an address controller.
16. The method according to claim 15 , wherein at least two larger RAM blocks share the same address controller.
17. The method according to claim 14 , wherein the address controller is a counter that is incremented by one after every stage of decoding the LPDCCC.
18. The method according to claim 13 , wherein the CNP is a look-up table tree.