Minghai Qin

Joint Source-Channel Decoding of Polar Codes for Language-Based Sources

We propose a joint list decoder and language decoder that exploits the redundancy of language- based sources during polar decoding. By judging the validity of decoded words in the decoded sequence with the help of a dictionary, the polar list decoder constantly detects erroneous paths after the decoding of every few bits. This path-pruning technique based on joint decoding has advantages over stand-alone polar list decoding in that most decoding errors in early stages are corrected. We show that if the language structure can be modeled as erasure correcting outer block codes, the rate of inner polar code can be increased while still guaranteeing a vanishing probability of error. To facilitate practical joint decoding, we first propose a construction of a dynamic dictionary using a trie and show an efficient way to trace the dictionary during decoding. Then we propose a joint decoding scheme for polar codes taking into account both information from the channel and the source. The proposed scheme has the same decoding complexity as the list decoding of polar codes. A list-size adaptive joint decoding is further implemented to largely reduce the decoding complexity. Simulation results show that the joint decoding schemes outperform stand-alone polar codes with CRC-aided successive cancellation list decoding by over 0.6 dB.

Balanced codes for data retention of multi-level flash memories with fast page read

Flash memories use the amount of charge (e.g., electrons) trapped in floating gate transistors to represent data. Charge leakage will cause data retention problem by unidirectionally shifting the cell-level distribution. Balanced codes are an effective means to adjust read thresholds adaptively and tolerate the charge leakage under unknown retention environments. For multi-level cell (MLC) flash memories with 4 levels, 2 binary logical pages are mapped to one quaternary physical page. In order to achieve a small read latency, one (or two, respectively) read threshold(s) is (are) applied to the physical page to retrieve data for the least significant bits (LSBs) (or most significant bits (MSBs), respectively). In this paper, we propose two coding schemes that use balanced codes for multi-level flash memories, which tolerate charge leakage and provide fast page read simultaneously. We prove that our coding schemes asymptotically bring no rate penalty (2 bits/cell for quaternary cells). We establish a model of the programming, retention, and read for MLC, then theoretically calculate the raw bit-error-rate (RBER) and page-error-rate (PER), and compare the performance of read threshold adjustments based on balanced codes and the conventional scheme that estimates the cell level drift. It is shown that for a fixed PER, the requirement for error correction codes (ECCs) when using balanced codes is much less than that of using conventional schemes, which has the benefits of ECC rates, decoding delay, and decoding complexity.

Enhancing the Expected Lifetime of NAND Flash by Short

Write-once memory (WOM) codes can be used to enhance the lifetime of multi-level flash memories by constraining unidirectional changes of cell levels. A

JOINT DECODING OF REWRITING NVM ERROR SECTORS

q

Training Recurrent Neural Networks against Noisy Computations during Inference.

-ary WOM code on integer lattices can be defined by a message decoding function and an updating function. Instead of focusing on the worst-case performance, i.e., the guaranteed number of writes of WOM codes, we study the average number of writes that can be successfully performed, assuming that the input alphabet is the same on each write. We model the updating of messages in a WOM as a Markov process on lattices, and present techniques to evaluate the average number of writes. Several code constructions are compared. A greedy algorithm is presented to obtain the optimal updating function of a given decoding function.