Refine
Document Type
- Conference Proceeding (9)
- Article (6)
Language
- English (15) (remove)
Keywords
- Binary codes (1)
- Block codes (1)
- Capacity (1)
- Channel capacity (1)
- Channel coding (1)
- Channel estimation (2)
- Computational complexity (1)
- Concatenated codes (2)
- Data retention time (1)
- Decoding (1)
Institute
Non-volatile NAND flash memories store information as an electrical charge. Different read reference voltages are applied to read the data. However, the threshold voltage distributions vary due to aging effects like program erase cycling and data retention time. It is necessary to adapt the read reference voltages for different life-cycle conditions to minimize the error probability during readout. In the past, methods based on pilot data or high-resolution threshold voltage histograms were proposed to estimate the changes in voltage distributions. In this work, we propose a machine learning approach with neural networks to estimate the read reference voltages. The proposed method utilizes sparse histogram data for the threshold voltage distributions. For reading the information from triple-level cell (TLC) memories, several read reference voltages are applied in sequence. We consider two histogram resolutions. The simplest histogram consists of the zero-and-one ratios for the hard decision read operation, whereas a higher resolution is obtained by considering the quantization levels for soft-input decoding. This approach does not require pilot data for the voltage adaptation. Furthermore, only a few measurements of extreme points of the threshold voltage distributions are required as training data. Measurements with different conditions verify the proposed approach. The resulting neural networks perform well under other life-cycle conditions.
Spatial modulation (SM) is a low-complexity multiple-input/multiple-output transmission technique that combines index modulation and quadrature amplitude modulation for wireless communications. In this work, we consider the problem of link adaption for generalized spatial modulation (GSM) systems that use multiple active transmit antennas simultaneously. Link adaption algorithms require a real-time estimation of the link quality of the time-variant communication channels, e.g., by means of estimating the mutual information. However, determining the mutual information of SM is challenging because no closed-form expressions have been found so far. Recently, multilayer feedforward neural networks were applied to compute the achievable rate of an index modulation link. However, only a small SM system with two transmit and two receive antennas was considered. In this work, we consider a similar approach but investigate larger GSM systems with multiple active antennas. We analyze the portions of mutual information related to antenna selection and the IQ modulation processes, which depend on the GSM variant and the signal constellation.
Reliability is a crucial aspect of non-volatile NAND flash memories, and it is essential to thoroughly analyze the channel to prevent errors and ensure accurate readout. Es-timating the read reference voltages (RRV s) is a significant challenge due to the multitude of physical effects involved. The question arises which features are useful and necessary for the RRV estimation. Various possible features require specialized hardware or specific readout techniques to be usable. In contrast we consider sparse histograms based on the decision thresholds for hard-input and soft-input decoding. These offer a distinct advantage as they are derived directly from the raw readout data without the need for decoding. This paper focuses on the information-theoretic study of different features, especially on the exploration of the mutual information (MI) between feature vector and RRV. In particular, we investigate the dependency of the MI on the resolution of the histograms. With respect to the RRV estimation, sparse histograms provide sufficient information for near-optimum estimation.
The growing error rates of triple-level cell (TLC) and quadruple-level cell (QLC) NAND flash memories have led to the application of error correction coding with soft-input decoding techniques in flash-based storage systems. Typically, flash memory is organized in pages where the individual bits per cell are assigned to different pages and different codewords of the error-correcting code. This page-wise encoding minimizes the read latency with hard-input decoding. To increase the decoding capability, soft-input decoding is used eventually due to the aging of the cells. This soft-decoding requires multiple read operations. Hence, the soft-read operations reduce the achievable throughput, and increase the read latency and power consumption. In this work, we investigate a different encoding and decoding approach that improves the error correction performance without increasing the number of reference voltages. We consider TLC and QLC flashes where all bits are jointly encoded using a Gray labeling. This cell-wise encoding improves the achievable channel capacity compared with independent page-wise encoding. Errors with cell-wise read operations typically result in a single erroneous bit per cell. We present a coding approach based on generalized concatenated codes that utilizes this property.
Mutual Information Analysis for Generalized Spatial Modulation Systems With Multilevel Coding
(2022)
Generalized Spatial Modulation (GSM) enables a trade-off between very high spectral efficiencies and low hardware costs for massive MIMO systems. This is achieved by transmitting information via the selection of active antennas from a set of available antennas besides the transmission of conventional data symbols. GSM systems have been investigated concerning various aspects like suitable signal constellations, efficient detection algorithms, hardware implementations, spatial precoding, and error control coding. On the other hand, determining the capacity of GSM is challenging because no closed-form expressions have been found so far. This paper investigates the mutual information for different GSM variants. We consider a multilevel coding approach, where the antenna selection and IQ modulation are encoded independently. Combined with multistage decoding, such an approach enables low-complexity capacity-achieving coded modulation. The influence of the data symbols on the mutual information is illuminated. We analyze the portions of mutual information related to antenna selection and the IQ modulation processes which depend on the GSM variant and the signal constellation. Moreover, the potential of spatial modulation for massive MIMO systems with many transmit antennas is investigated. Especially in systems with many transmit antennas much information can be conveyed by antenna selection.
Reed-Muller (RM) codes have recently regained some interest in the context of low latency communications and due to their relation to polar codes. RM codes can be constructed based on the Plotkin construction. In this work, we consider concatenated codes based on the Plotkin construction, where extended Bose-Chaudhuri-Hocquenghem (BCH) codes are used as component codes. This leads to improved code parameters compared to RM codes. Moreover, this construction is more flexible concerning the attainable code rates. Additionally, new soft-input decoding algorithms are proposed that exploit the recursive structure of the concatenation and the cyclic structure of the component codes. First, we consider the decoding of the cyclic component codes and propose a low complexity hybrid ordered statistics decoding algorithm. Next, this algorithm is applied to list decoding of the Plotkin construction. The proposed list decoding approach achieves near-maximum-likelihood performance for codes with medium lengths. The performance is comparable to state-of-the-art decoders, whereas the complexity is reduced.
Automotive computing applications like AI databases, ADAS, and advanced infotainment systems have a huge need for persistent memory. This trend requires NAND flash memories designed for extreme automotive environments. However, the error probability of NAND flash memories has increased in recent years due to higher memory density and production tolerances. Hence, strong error correction coding is needed to meet automotive storage requirements. Many errors can be corrected by soft decoding algorithms. However, soft decoding is very resource-intensive and should be avoided when possible. NAND flash memories are organized in pages, and the error correction codes are usually encoded page-wise to reduce the latency of random reads. This page-wise encoding does not reach the maximum achievable capacity. Reading soft information increases the channel capacity but at the cost of higher latency and power consumption. In this work, we consider cell-wise encoding, which also increases the capacity compared to page-wise encoding. We analyze the cell-wise processing of data in triple-level cell (TLC) NAND flash and show the performance gain when using Low-Density Parity-Check (LDPC) codes. In addition, we investigate a coding approach with page-wise encoding and cell-wise reading.
Large persistent memory is crucial for many applications in embedded systems and automotive computing like AI databases, ADAS, and cutting-edge infotainment systems. Such applications require reliable NAND flash memories made for harsh automotive conditions. However, due to high memory densities and production tolerances, the error probability of NAND flash memories has risen. As the number of program/erase cycles and the data retention times increase, non-volatile NAND flash memories' performance and dependability suffer. The read reference voltages of the flash cells vary due to these aging processes. In this work, we consider the issue of reference voltage adaption. The considered estimation procedure uses shallow neural networks to estimate the read reference voltages for different life-cycle conditions with the help of histogram measurements. We demonstrate that the training data for the neural networks can be enhanced by using shifted histograms, i.e., a training of the neural networks is possible based on a few measurements of some extreme points used as training data. The trained neural networks generalize well for other life-cycle conditions.
The performance and reliability of non-volatile NAND flash memories deteriorate as the number of program/erase cycles grows. The reliability also suffers from cell to cell interference, long data retention time, and read disturb. These processes effect the read threshold voltages. The aging of the cells causes voltage shifts which lead to high bit error rates (BER) with fixed pre-defined read thresholds. This work proposes two methods that aim on minimizing the BER by adjusting the read thresholds. Both methods utilize the number of errors detected in the codeword of an error correction code. It is demonstrated that the observed number of errors is a good measure for the voltage shifts and is utilized for the initial calibration of the read thresholds. The second approach is a gradual channel estimation method that utilizes the asymmetrical error probabilities for the one-to-zero and zero-to-one errors that are caused by threshold calibration errors. Both methods are investigated utilizing the mutual information between the optimal read voltage and the measured error values.
Numerical results obtained from flash measurements show that these methods reduce the BER of NAND flash memories significantly.
Error correction coding for optical communication and storage requires high rate codes that enable high data throughput and low residual errors. Recently, different concatenated coding schemes were proposed that are based on binary BCH codes with low error correcting capabilities. In this work, low-complexity hard- and soft-input decoding methods for such codes are investigated. We propose three concepts to reduce the complexity of the decoder. For the algebraic decoding we demonstrate that Peterson's algorithm can be more efficient than the Berlekamp-Massey algorithm for single, double, and triple error correcting BCH codes. We propose an inversion-less version of Peterson's algorithm and a corresponding decoding architecture. Furthermore, we propose a decoding approach that combines algebraic hard-input decoding with soft-input bit-flipping decoding. An acceptance criterion is utilized to determine the reliability of the estimated codewords. For many received codewords the stopping criterion indicates that the hard-decoding result is sufficiently reliable, and the costly soft-input decoding can be omitted. To reduce the memory size for the soft-values, we propose a bit-flipping decoder that stores only the positions and soft-values of a small number of code symbols. This method significantly reduces the memory requirements and has little adverse effect on the decoding performance.