pith. sign in

arxiv: 1907.03938 · v1 · pith:TWP7VQ3Inew · submitted 2019-07-09 · 💻 cs.IT · math.IT

Deep Learning-Aided Dynamic Read Thresholds Design For Multi-Level-Cell Flash Memories

Zhen Mei , Kui Cai , Xuan He This is my paper

Pith reviewed 2026-05-25 00:31 UTC · model grok-4.3

classification 💻 cs.IT math.IT
keywords flash memorydynamic read thresholdsrecurrent neural networkMLCLDPC codesdata retention noisesoft-decision decodingdensity evolution
0
0 comments X

The pith

An RNN detector can derive dynamic read thresholds for MLC flash memory that work without any prior channel knowledge.

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

NAND flash cells experience shifting noise and unknown offsets from data retention that fixed thresholds cannot handle. The paper establishes that a recurrent neural network can detect stored symbols in multi-level cells directly from raw reads. To limit the latency and power cost of the RNN, its outputs are used only occasionally to compute thresholds for a standard detector that runs the rest of the time. These thresholds further support soft decoding by generating additional levels and integer reliability values. Density evolution paired with differential evolution then tunes the thresholds for LDPC-coded channels.

Core claim

The paper claims that read thresholds derived from periodic outputs of an RNN detector yield accurate symbol detection in MLC flash memories even when the channel offset remains unknown, and that the same thresholds can be extended to produce soft information suitable for LDPC decoding after optimization by density evolution and differential evolution.

What carries the argument

The RNN-aided (RNNA) dynamic threshold detector, which extracts hard-decision thresholds from the outputs of a recurrent neural network detector activated only when the system is idle.

If this is right

  • The RNNA thresholds improve detection performance over fixed thresholds for both uncoded and LDPC-coded flash channels.
  • Additional read thresholds and integer reliability mappings can be generated from the hard-decision thresholds to supply soft information to the decoder.
  • Density evolution combined with differential evolution produces optimized thresholds for LDPC-coded channels.
  • The RNN detector needs activation only periodically, limiting its latency and power impact during normal operation.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • The periodic-activation pattern could allow the RNN component to be implemented as a low-duty-cycle coprocessor rather than a continuous pipeline.
  • The same derivation approach might adapt thresholds across different flash generations or operating temperatures without retraining.
  • Integer reliability mappings derived this way could reduce the precision required in the soft decoder hardware.

Load-bearing premise

Outputs from the RNN detector contain enough information to set read thresholds that remain accurate even though the channel offset and noise statistics are unknown.

What would settle it

Apply the derived RNNA thresholds to a sequence of MLC flash reads with increasing retention time and measure whether the resulting bit-error rate stays below that of a conventional fixed-threshold detector.

Figures

Figures reproduced from arXiv: 1907.03938 by Kui Cai, Xuan He, Zhen Mei.

Figure 1
Figure 1. Figure 1: Threshold voltage distributions of MLC flash memory c [PITH_FULL_IMAGE:figures/full_fig_p003_1.png] view at source ↗
Figure 3
Figure 3. Figure 3: The training SER of the RNN detector for each epoch at P [PITH_FULL_IMAGE:figures/full_fig_p004_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: Defining the non-uniform read thresholds [PITH_FULL_IMAGE:figures/full_fig_p006_4.png] view at source ↗
Figure 5
Figure 5. Figure 5: The MI with the MMI quantizer and RNNA quantizer over d [PITH_FULL_IMAGE:figures/full_fig_p008_5.png] view at source ↗
Figure 7
Figure 7. Figure 7: BER of the RNN detector and RNNA dynamic threshold det [PITH_FULL_IMAGE:figures/full_fig_p008_7.png] view at source ↗
Figure 10
Figure 10. Figure 10: BERs of LDPC codes with the RNNA quantizer, trained w [PITH_FULL_IMAGE:figures/full_fig_p009_10.png] view at source ↗
Figure 9
Figure 9. Figure 9: BER comparison of different detectors. The RNNA dyna [PITH_FULL_IMAGE:figures/full_fig_p009_9.png] view at source ↗
Figure 11
Figure 11. Figure 11: BERs of LDPC codes with the RNNA quantizer, trained w [PITH_FULL_IMAGE:figures/full_fig_p010_11.png] view at source ↗
read the original abstract

The practical NAND flash memory suffers from various non-stationary noises that are difficult to be predicted. Furthermore, the data retention noise induced channel offset is unknown during the readback process. This severely affects the data recovery from the memory cell. In this paper, we first propose a novel recurrent neural network (RNN)-based detector to effectively detect the data symbols stored in the multi-level-cell (MLC) flash memory without any prior knowledge of the channel. However, compared with the conventional threshold detector, the proposed RNN detector introduces much longer read latency and more power consumption. To tackle this problem, we further propose an RNN-aided (RNNA) dynamic threshold detector, whose detection thresholds can be derived based on the outputs of the RNN detector. We thus only need to activate the RNN detector periodically when the system is idle. Moreover, to enable soft-decision decoding of error-correction codes, we first show how to obtain more read thresholds based on the hard-decision read thresholds derived from the RNN detector. We then propose integer-based reliability mappings based on the designed read thresholds, which can generate the soft information of the channel. Finally, we propose to apply density evolution (DE) combined with differential evolution algorithm to optimize the read thresholds for LDPC coded flash memory channels. Computer simulation results demonstrate the effectiveness of our RNNA dynamic read thresholds design, for both the uncoded and LDPC-coded flash memory channels, without any prior knowledge of the channel.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Referee Report

2 major / 2 minor

Summary. The paper proposes an RNN-based detector for symbol detection in MLC NAND flash that claims to operate without prior channel knowledge, followed by an RNNA dynamic threshold detector that derives read thresholds from periodic RNN activations to reduce latency. It further develops additional thresholds and integer reliability mappings for soft-decision LDPC decoding, optimizes thresholds via density evolution combined with differential evolution, and presents simulations claiming effectiveness for both uncoded and coded channels without prior knowledge.

Significance. If the no-prior-knowledge claim can be substantiated with explicit training details, the work could provide a useful adaptive approach for handling retention-induced offsets and non-stationary noise in flash memories. The periodic activation strategy and extension to soft information are practical ideas, and the use of DE for LDPC threshold optimization is a standard but well-motivated step.

major comments (2)
  1. [Abstract] Abstract: The central claim that the RNN detector (and thus the derived RNNA thresholds) functions 'without any prior knowledge of the channel' is load-bearing for the paper's contribution, yet the manuscript supplies no information on how the RNN is trained (e.g., whether supervised training uses channel-generated labeled pairs, an assumed model, or unsupervised methods). Standard RNN training for detection requires data that encodes the very noise and offset statistics the method claims to handle blindly, creating an unresolved dependency.
  2. [Abstract] Abstract (simulation results paragraph): The effectiveness claims for both uncoded and LDPC-coded cases rest on unspecified computer simulations with no reported details on training data generation, baseline comparators (e.g., conventional threshold detectors or model-based methods), error bars, or exclusion criteria. This prevents verification that the reported gains are attributable to the no-prior-knowledge property rather than implicit channel information in the training set.
minor comments (2)
  1. The abstract and introduction would benefit from a brief statement of the RNN architecture (number of layers, activation functions, input/output dimensions) to allow readers to assess computational overhead.
  2. Notation for the reliability mapping and integer-based soft information should be defined explicitly when first introduced, as the transition from hard-decision thresholds to soft values is central to the LDPC extension.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the constructive comments highlighting the need for greater clarity on the training procedure and simulation details. We will revise the manuscript to address these points explicitly.

read point-by-point responses

  1. Referee: [Abstract] Abstract: The central claim that the RNN detector (and thus the derived RNNA thresholds) functions 'without any prior knowledge of the channel' is load-bearing for the paper's contribution, yet the manuscript supplies no information on how the RNN is trained (e.g., whether supervised training uses channel-generated labeled pairs, an assumed model, or unsupervised methods). Standard RNN training for detection requires data that encodes the very noise and offset statistics the method claims to handle blindly, creating an unresolved dependency.

    Authors: We agree that the manuscript does not provide sufficient detail on RNN training, which weakens the substantiation of the no-prior-knowledge claim as presented. In the revision we will add an explicit description of the training process, noting that supervised training uses labeled pairs generated from a simulated MLC channel model that incorporates representative retention and noise statistics. The trained model is then deployed without requiring runtime knowledge of specific channel parameters. We will also clarify that this offline training step is distinct from online channel estimation required by conventional model-based detectors. revision: yes

  2. Referee: [Abstract] Abstract (simulation results paragraph): The effectiveness claims for both uncoded and LDPC-coded cases rest on unspecified computer simulations with no reported details on training data generation, baseline comparators (e.g., conventional threshold detectors or model-based methods), error bars, or exclusion criteria. This prevents verification that the reported gains are attributable to the no-prior-knowledge property rather than implicit channel information in the training set.

    Authors: We concur that the simulation description is incomplete and prevents proper evaluation. The revised manuscript will include details on training-data generation (using the same general channel model employed for RNN training), the specific baseline detectors compared, the number of Monte Carlo trials, and reporting of variability measures such as error bars. These additions will allow readers to assess whether gains arise from the adaptive, knowledge-free operation at inference time. revision: yes

Circularity Check

0 steps flagged

No circularity; RNN training and DE optimization are external to the claimed derivation.

full rationale

The abstract and described method introduce an RNN detector trained on external data, derive thresholds from its outputs, map to soft information, and optimize via standard density evolution plus differential evolution. No equations, self-citations, or steps are quoted that reduce a claimed prediction or result to a fitted parameter or prior self-result by construction. The 'without prior knowledge' assertion is presented as validated by simulation rather than derived tautologically from the inputs. This is the normal case of a self-contained proposal relying on data-driven training and established algorithms.

Axiom & Free-Parameter Ledger

0 free parameters · 0 axioms · 0 invented entities

Abstract supplies no equations, fitted constants, or explicit axioms; the approach implicitly assumes standard RNN training converges to useful detectors and that DE accurately models the coded channel, but none are stated or quantified.

pith-pipeline@v0.9.0 · 5791 in / 1059 out tokens · 30210 ms · 2026-05-25T00:31:51.569745+00:00 · methodology

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Lean theorems connected to this paper

Citations machine-checked in the Pith Canon. Every link opens the source theorem in the public Lean library.

  • IndisputableMonolith/Cost/FunctionalEquation.lean washburn_uniqueness_aczel unclear
    ?
    unclear

    Relation between the paper passage and the cited Recognition theorem.

    we first propose a novel recurrent neural network (RNN)-based detector to effectively detect the data symbols stored in the multi-level-cell (MLC) flash memory without any prior knowledge of the channel... RNN-aided (RNNA) dynamic threshold detector, whose detection thresholds can be derived based on the outputs of the RNN detector

  • IndisputableMonolith/Foundation/ArithmeticFromLogic.lean LogicNat recovery unclear
    ?
    unclear

    Relation between the paper passage and the cited Recognition theorem.

    we propose to apply density evolution (DE) combined with differential evolution algorithm to optimize the read thresholds for LDPC coded flash memory channels

What do these tags mean?
matches
The paper's claim is directly supported by a theorem in the formal canon.
supports
The theorem supports part of the paper's argument, but the paper may add assumptions or extra steps.
extends
The paper goes beyond the formal theorem; the theorem is a base layer rather than the whole result.
uses
The paper appears to rely on the theorem as machinery.
contradicts
The paper's claim conflicts with a theorem or certificate in the canon.
unclear
Pith found a possible connection, but the passage is too broad, indirect, or ambiguous to say the theorem truly supports the claim.

Reference graph

Works this paper leans on

44 extracted references · 44 canonical work pages

  1. [1]

    Err or characterization, mitigation, and recovery in flash-memor y- based solid-state drives,

    Y . Cai, S. Ghose, E. F. Haratsch, Y . Luo, and O. Mutlu, “Err or characterization, mitigation, and recovery in flash-memor y- based solid-state drives,” Proceedings of the IEEE , vol. 105, no. 9, pp. 1666–1704, 2017

    work page 2017

  2. [2]

    Error patter ns in mlc nand flash memory: Measurement, characterization, an d analysis,

    Y . Cai, E. F. Haratsch, O. Mutlu, and K. Mai, “Error patter ns in mlc nand flash memory: Measurement, characterization, an d analysis,” in Proc. DATE, 2012

    work page 2012

  3. [3]

    6.4 gb/s multi-thr eaded bch encoder and decoder for multi-channel ssd controllers,

    Y . Lee, H. Y oo, I. Y oo, and I.-C. Park, “6.4 gb/s multi-thr eaded bch encoder and decoder for multi-channel ssd controllers, ” in Proc. ISSCC , 2012

    work page 2012

  4. [4]

    High-throughput and low-complexity BCH decoding architecture for solid-state drives,

    ——, “High-throughput and low-complexity BCH decoding architecture for solid-state drives,” IEEE Trans. V ery Large Scale Integr . (VLSI) Syst., vol. 22, no. 5, pp. 1183–1187, 2014

    work page 2014

  5. [5]

    LDPC-in-SSD: Making advanced error correction codes work effectively in solid state drives,

    K. Zhao, W. Zhao, H. Sun, X. Zhang, N. Zheng, and T. Zhang, “LDPC-in-SSD: Making advanced error correction codes work effectively in solid state drives,” in Proc. F AST, 2013

    work page 2013

  6. [6]

    On the use of soft-decision error-correction codes in nand flash memory,

    G. Dong, N. Xie, and T. Zhang, “On the use of soft-decision error-correction codes in nand flash memory,” IEEE Trans. Circuits Syst. I, Reg. Papers , vol. 58, no. 2, pp. 429–439, 2011

    work page 2011

  7. [7]

    Enhanced precision through multiple reads for LDPC decoding in flash memories,

    J. Wang, K. V akilinia, T.-Y . Chen, T. Courtade, G. Dong, T. Zhang, H. Shankar, and R. Wesel, “Enhanced precision through multiple reads for LDPC decoding in flash memories,” IEEE J. Sel. Areas Commun. , vol. 32, no. 5, pp. 880–891, May 2014

    work page 2014

  8. [8]

    Adaptive read thresholds for nand flash,

    B. Peleato, R. Agarwal, J. M. Cioffi, M. Qin, and P . H. Siegel, “Adaptive read thresholds for nand flash,” IEEE Trans. Commun., vol. 63, no. 9, pp. 3069–3081, 2015

    work page 2015

  9. [9]

    Read and write voltage signal optimization for multi-level-cell (mlc) na nd flash memory,

    C. A. Aslam, Y . L. Guan, and K. Cai, “Read and write voltage signal optimization for multi-level-cell (mlc) na nd flash memory,” IEEE Trans. Commun., vol. 64, no. 4, pp. 1613–1623, 2016

    work page 2016

  10. [10]

    Information theoretic bound s based channel quantization design for emerging memories,

    Z. Mei, K. Cai, and L. Shi, “Information theoretic bound s based channel quantization design for emerging memories,” in Proc. IEEE ITW , Nov. 2018

    work page 2018

  11. [11]

    Data retention in MLC NAND flash memory: Characterization, op- timization, and recovery,

    Y . Cai, Y . Luo, E. F. Haratsch, K. Mai, and O. Mutlu, “Data retention in MLC NAND flash memory: Characterization, op- timization, and recovery,” in Proc. IEEE HPCA , 2015

    work page 2015

  12. [12]

    Rank m od- ulation for flash memories,

    A. Jiang, R. Mateescu, M. Schwartz, and J. Bruck, “Rank m od- ulation for flash memories,” IEEE Trans. Inf. Theory , vol. 55, no. 6, pp. 2659–2673, 2009

    work page 2009

  13. [13]

    Design of so me new efficient balanced codes,

    L. G. Tallini, R. M. Capocelli, and B. Bose, “Design of so me new efficient balanced codes,” IEEE Trans. Inf. Theory , vol. 42, no. 3, pp. 790–802, 1996

    work page 1996

  14. [14]

    Composition check codes,

    K. A. S. Immink and K. Cai, “Composition check codes,” IEEE Trans. Inf. Theory , vol. 64, no. 1, pp. 249–256, 2017

    work page 2017

  15. [15]

    Error-correcting sche mes with dynamic thresholds in nonvolatile memories,

    H. Zhou, A. Jiang, and J. Bruck, “Error-correcting sche mes with dynamic thresholds in nonvolatile memories,” in Proc. IEEE ISIT, 2011

    work page 2011

  16. [16]

    Dynamic threshold schemes for multi-level non-volatile memories,

    F. Sala, R. Gabrys, and L. Dolecek, “Dynamic threshold schemes for multi-level non-volatile memories,” IEEE Trans. Commun., vol. 61, no. 7, pp. 2624–2634, 2013

    work page 2013

  17. [17]

    An introduction to deep learnin g for the physical layer,

    T. OShea and J. Hoydis, “An introduction to deep learnin g for the physical layer,” IEEE Trans. Cogn. Commun. Netw , vol. 3, no. 4, pp. 563–575, 2017

    work page 2017

  18. [18]

    On d eep learning-based channel decoding,

    T. Gruber, S. Cammerer, J. Hoydis, and S. ten Brink, “On d eep learning-based channel decoding,” in In Proc. IEEE CISS , Mar. 2017

    work page 2017

  19. [19]

    Enabling nand flash memory use soft-decision error correction codes at minimal read la tency overhead,

    G. Dong, N. Xie, and T. Zhang, “Enabling nand flash memory use soft-decision error correction codes at minimal read la tency overhead,” IEEE Trans. Circuits Syst. I, Reg. Papers , vol. 60, no. 9, pp. 2412–2421, 2013

    work page 2013

  20. [20]

    Histogram-based fl ash channel estimation,

    H. Wang, T.-Y . Chen, and R. D. Wesel, “Histogram-based fl ash channel estimation,” in Proc. IEEE ICC , Jun. 2015

    work page 2015

  21. [21]

    A double-leve l- v/sub th/select gate array architecture for multilevel nan d flash memories,

    K. Takeuchi, T. Tanaka, and H. Nakamura, “A double-leve l- v/sub th/select gate array architecture for multilevel nan d flash memories,” IEEE J. Solid-State Circuits , vol. 31, no. 4, pp. 602– 609, 1996

    work page 1996

  22. [22]

    Effects of floatin g-gate interference on NAND flash memory cell operation,

    J.-D. Lee, S.-H. Hur, and J.-D. Choi, “Effects of floatin g-gate interference on NAND flash memory cell operation,” IEEE Electron Device Lett. , vol. 23, no. 5, pp. 264–266, 2002

    work page 2002

  23. [23]

    Using lifetime-aware pro gres- sive programming to improve SLC NAND flash memory write endurance,

    G. Dong, Y . Pan, and T. Zhang, “Using lifetime-aware pro gres- sive programming to improve SLC NAND flash memory write endurance,” IEEE Transactions on V ery Large Scale Integration (VLSI) Systems , vol. 22, no. 6, pp. 1270–1280, 2014

    work page 2014

  24. [24]

    First evidence for injection statistics accuracy limitations in NAND flash con stant- current fowler-nordheim programming,

    C. M. Compagnoni, A. Spinelli, R. Gusmeroli, A. L. Lacai ta, S. Beltrami, A. Ghetti, and A. Visconti, “First evidence for injection statistics accuracy limitations in NAND flash con stant- current fowler-nordheim programming,” in Proc. IEEE Int. Electron Devices Meeting (IEDM) , 2007

    work page 2007

  25. [25]

    Flash EEPROM threshold in- stabilities due to charge trapping during program/erase cy cling,

    N. Mielke, H. Belgal, I. Kalastirsky, P . Kalavade, A. Ku rtz, Q. Meng, N. Righos, and J. Wu, “Flash EEPROM threshold in- stabilities due to charge trapping during program/erase cy cling,” IEEE Trans. Device Mater . Rel. , vol. 4, no. 3, pp. 335–344, 2004

    work page 2004

  26. [26]

    Random telegraph noise effect on the pro- grammed threshold-voltage distribution of flash memories,

    C. M. Compagnoni, M. Ghidotti, A. L. Lacaita, A. S. Spine lli, and A. Visconti, “Random telegraph noise effect on the pro- grammed threshold-voltage distribution of flash memories, ” IEEE Electron Device Lett. , vol. 30, no. 9, pp. 984–986, 2009

    work page 2009

  27. [27]

    Decision-directed retention-failure recovery with channel update for MLC NAN D flash memory,

    C. A. Aslam, Y . L. Guan, and K. Cai, “Decision-directed retention-failure recovery with channel update for MLC NAN D flash memory,” IEEE Trans. Circuits Syst. I, Reg. Papers , vol. 65, no. 1, pp. 353–365, 2018

    work page 2018

  28. [28]

    Increasi ng flash memory lifetime by dynamic voltage allocation for con- stant mutual information,

    T.-Y . Chen, A. R. Williamson, and R. D. Wesel, “Increasi ng flash memory lifetime by dynamic voltage allocation for con- stant mutual information,” in Proc. Inf. Theory Appl. W orkshop, Feb. 2014

    work page 2014

  29. [29]

    Estimating information-theoretical nand flash memory storage capacit y and its implication to memory system design space exploration,

    G. Dong, Y . Pan, N. Xie, C. V aranasi, and T. Zhang, “Estimating information-theoretical nand flash memory storage capacit y and its implication to memory system design space exploration, ” IEEE Trans. VLSI Syst. , vol. 20, no. 9, pp. 1705–1714, 2012

    work page 2012

  30. [30]

    Using data postcompensati on and predistortion to tolerate cell-to-cell interference i n MLC NAND flash memory,

    G. Dong, S. Li, and T. Zhang, “Using data postcompensati on and predistortion to tolerate cell-to-cell interference i n MLC NAND flash memory,” IEEE Trans. Circuits Syst. I, Reg. Papers, vol. 57, no. 10, pp. 2718–2728, 2010

    work page 2010

  31. [31]

    Threshold voltage distribution in mlc nand flash memory: Characteriza tion, analysis, and modeling,

    Y . Cai, E. F. Haratsch, O. Mutlu, and K. Mai, “Threshold voltage distribution in mlc nand flash memory: Characteriza tion, analysis, and modeling,” in Proc. DATE, Mar. 2013

    work page 2013

  32. [32]

    Estimation of nand flash memory threshold voltage distribution for optimum soft-decision error correction,

    D.-h. Lee and W. Sung, “Estimation of nand flash memory threshold voltage distribution for optimum soft-decision error correction,” IEEE Trans. Signal Process. , vol. 61, no. 2, pp. 440–449, 2013

    work page 2013

  33. [33]

    Goodfellow, Y

    I. Goodfellow, Y . Bengio, A. Courville, and Y . Bengio, Deep learning. MIT press Cambridge, 2016, vol. 1

    work page 2016

  34. [34]

    T. H. Cormen, C. E. Leiserson, R. L. Rivest, and C. Stein, Introduction to Algorithms, Third Edition , 3rd ed. The MIT Press, 2009

    work page 2009

  35. [35]

    Dynamic programming for discrete memoryless channel quantization,

    X. He, K. Cai, W. Song, and Z. Mei, “Dynamic programming for discrete memoryless channel quantization,” arXiv preprint arXiv:1901.01659, 2019

    work page arXiv 1901

  36. [36]

    Boyd and L

    S. Boyd and L. V andenberghe, Convex optimization . Cam- bridge university press, 2004

    work page 2004

  37. [37]

    Comparisons betwee n reliability-based iterative min-sum and majority-logic d ecoding algorithms for ldpc codes,

    H. Chen, K. Zhang, X. Ma, and B. Bai, “Comparisons betwee n reliability-based iterative min-sum and majority-logic d ecoding algorithms for ldpc codes,” IEEE Trans. Commun. , vol. 59, no. 7, pp. 1766–1771, 2011

    work page 2011

  38. [38]

    Channel capacity and soft- decision decoding of LDPC codes for spin-torque transfer ma g- netic random access memory (STT-MRAM),

    K. Cai, Z. Qin, and B. Chen, “Channel capacity and soft- decision decoding of LDPC codes for spin-torque transfer ma g- netic random access memory (STT-MRAM),” in Proc. IEEE ICNC, Jan. 2013

    work page 2013

  39. [39]

    Richardson and R

    T. Richardson and R. Urbanke, Modern coding theory . Cam- bridge university press, 2008

    work page 2008

  40. [40]

    Reduced-complexity decoding of ldpc codes,

    J. Chen, A. Dholakia, E. Eleftheriou, M. P . Fossorier, a nd X.-Y . Hu, “Reduced-complexity decoding of ldpc codes,” IEEE trans. Commun., vol. 53, no. 8, pp. 1288–1299, 2005

    work page 2005

  41. [41]

    On the design of low-density parity-check codes within 0.0 045 db of the shannon limit,

    S.-Y . Chung, G. D. Forney, T. J. Richardson, and R. Urban ke, “On the design of low-density parity-check codes within 0.0 045 db of the shannon limit,” IEEE Commun. lett. , vol. 5, no. 2, pp. 58–60, 2001

    work page 2001

  42. [42]

    Differential evolution–a simpl e and effi- cient heuristic for global optimization over continuous sp aces,

    R. Storn and K. Price, “Differential evolution–a simpl e and effi- cient heuristic for global optimization over continuous sp aces,” Journal of global optimization , vol. 11, no. 4, pp. 341–359, 1997

    work page 1997

  43. [43]

    Chollet, “keras,” https://github.com/keras-team /keras, 2015

    F. Chollet, “keras,” https://github.com/keras-team /keras, 2015

    work page 2015

  44. [44]

    Tensorflow: Large-scale machine learn- ing on heterogeneous systems,

    M. Abadi et al. , “Tensorflow: Large-scale machine learn- ing on heterogeneous systems,” 2015. [Online]. Available: http://tensorflow.org/

    work page 2015