Masaki HAMAMOTO Hiroyuki NAMBA Masashi EGI
Explainable artificial intelligence (AI) technology enables us to quantitatively analyze the whole prediction logic of AI as a global explanation. However, unwanted relationships learned by AI due to data sparsity, high dimensionality, and noise are also visualized in the explanation, which deteriorates confidence in the AI. Thus, methods for correcting those unwanted relationships in explanation has been developed. However, since these methods are applicable only to differentiable machine learning (ML) models but not to non-differentiable models such as tree-based models, they are insufficient for covering a wide range of ML technology. Since these methods also require re-training of the model for correcting its explanation (i.e., in-processing method), they cannot be applied to black-box models provided by third parties. Therefore, we propose a method called ensemble-based explanation correction (EBEC) as a post-processing method for correcting the global explanation of a prediction model in a model-agnostic manner by using the Rashomon effect of statistics. We evaluated the performance of EBEC with three different tasks and analyzed its function in more detail. The evaluation results indicate that EBEC can correct global explanation of the model so that the explanation aligns with the domain knowledge given by the user while maintaining its accuracy. EBEC can be extended in various ways and combined with any method to improve correction performance since it is a post-processing-type correction method. Hence, EBEC would contribute to high-productivity ML modeling as a new type of explanation-correction method.
An adaptive bit allocation scheme for zero-forcing (ZF) Tomlinson-Harashima precoding (THP) is proposed. The ZF-THP enables us to achieve feasible bit error rate (BER) performance when appropriate substream permutations are installed at the transmitter. In this study, the number of bits in each substream is adaptively allocated to minimize the average BER in fading environments. Numerical examples are provided to compare the proposed method with eigenbeam space division multiplexing (E-SDM) method.
Yosuke IIJIMA Yasushi YUMINAKA
The growing demand for high-speed data communication has continued to meet the need for ever-increasing I/O bandwidth in recent VLSI systems. However, signal integrity issues, such as intersymbol interference (ISI) and reflections, make the channel band-limited at high-speed data rates. We propose high-speed data transmission techniques for VLSI systems using Tomlinson-Harashima precoding (THP). Because THP can eliminate ISI by inverting the characteristics of channels with limited peak and average power at the transmitter, it is suitable for implementing advanced low-voltage and high-speed VLSI systems. This paper presents a novel double-rate THP equalization technique especially intended for multi-valued data transmission to further improve THP performance. Simulation and measurement results show that the proposed THP equalization with a double sampling rate can enhance the data transition time and, therefore, improve the eye opening.
Shohei YOSHIOKA Shinya KUMAGAI Fumiyuki ADACHI
Nonlinear precoding improves the downlink bit error rate (BER) performance of multi-user multiple-input multiple-output (MU-MIMO). Broadband single-carrier (SC) block transmission can improve the capability that nonlinear precoding reduces BER, as it provides frequency diversity gain. This paper considers Tomlinson-Harashima precoding (THP) as a nonlinear precoding scheme for SC-MU-MIMO downlink. In the SC-MU-MIMO downlink with frequency-domain THP proposed by Degen and Rrühl (called SC-FDTHP), the inter-symbol interference (ISI) is suppressed by transmit frequency-domain equalization (FDE) after suppressing the inter-user interference (IUI) by frequency-domain THP. Transmit FDE increases the signal variance, hence transmission performance improvement is limited. In this paper, we propose a new SC-MU-MIMO downlink with time-domain THP which can pre-remove both ISI and IUI (called SC-TDTHP) if perfect channel state information (CSI) is available. Modulo operation in THP suppresses the signal variance increase caused by ISI and IUI pre-removal, and hence the transmission quality improves. For further performance improvement, vector perturbation is introduced to SC-TDTHP (called SC-TDTHP w/VP). Computer simulation shows that SC-TDTHP achieves better BER performance than SC-FDTHP and that SC-TDTHP w/VP offers further improvement in BER performance over SC-MU-MIMO with VP (called SC-VP). Computational complexity is also compared and it is showed that SC-TDTHP and SC-TDTHP w/VP incur higher computational complexity than SC-FDTHP but lower than SC-VP.
In this paper, we propose a zero-forcing (ZF) Tomlinson-Harashima precoding (THP) with substream permutations based on the bit rate maximization for single-user MIMO (SU-MIMO) systems. We study the effect of substream permutations on the ZF-THP SU-MIMO systems, when the mean squared error (MSE) and the bit rate are adopted for the selection of the permutation matrix as criteria. Based on our analysis, we propose a method to increase the bit rate by substream permutations, and derive QR and Cholesky decomposition-based algorithms which realize the proposed method. Furthermore, to improve the error rate performance, we apply zero transmission to subchannels with low signal-to-noise ratios. Numerical examples are provided to demonstrate the effectiveness of the proposed THP MIMO system.
Yosuke IIJIMA Yuuki TAKADA Yasushi YUMINAKA
The data rate of VLSI interconnections has been increasing according to the demand for high-speed operation of semiconductors such as CPUs. To realize high performance VLSI systems, high-speed data communication has become an important factor. However, at high-speed data rates, it is difficult to achieve accurate communication without bit errors because of inter-symbol interference (ISI). This paper presents high-speed data communication techniques for VLSI systems using Tomlinson-Harashima Precoding (THP). Since THP can eliminate the ISI with limiting average and peak power of transmitter signaling, THP is suitable for implementing advanced low-voltage VLSI systems. In this paper, 4-PAM (Pulse amplitude modulation) with THP has been employed to achieve high-speed data communication in VLSI systems. Simulation results show that THP can remove the ISI without increasing peak and average power of a transmitter. Moreover, simulation results clarify that multiple-valued data communication is very effective to reduce implementation costs for realizing high-speed serial links.
Xunyong ZHANG Chen HE Lingge JIANG
In this paper, an effective per-antenna successive signal-to-leakage-plus-noise-ratio (PA-SSLNR) based precoding is proposed for multi-user multiple-input multiple-output (MIMO) broadcast channel. The signal-to-leakage-plus-noise-ratio (SLNR) of per-antenna is calculated only using the unknown leakages and the known leakages are cancelled at the transmit side by Tomlinson-Harashima Precoding (THP). The proposed scheme is different from per-user SSLNR. It does not need QR decomposition. The proposed precoding scheme is further improved by ordering antennas. Simulation results show that the proposed schemes exhibit a considerable bit error rate (BER) improvement over conventional SLNR scheme.
Kosuke SHIMAZAKI Shingo YOSHIZAWA Yasuyuki HATAKAWA Tomoko MATSUMOTO Satoshi KONISHI Yoshikazu MIYANAGA
This paper presents a VLSI design of a Tomlinson-Harashima (TH) precoder for multi-user MIMO (MU-MIMO) systems. The TH precoder consists of LQ decomposition (LQD), interference cancellation (IC), and weight coefficient multiplication (WCM) units. The LQ decomposition unit is based on an application specific instruction-set processor (ASIP) architecture with floating-point arithmetic for high accuracy operations. In the IC and WCM units with fixed-point arithmetic, the proposed architecture uses an arrayed pipeline structure to shorten a circuit critical path delay. The implementation result shows that the proposed architecture reduces circuit area and power consumption by 11% and 15%, respectively.
Shogo FUJITA Leonardo LANANTE Jr. Yuhei NAGAO Masayuki KUROSAKI Hiroshi OCHI
In this paper, we propose a modified Tomlinson Harashima precoding (THP) method with less increase of computational complexity for the multi-user MIMO downlink system. The proposed THP scheme minimizes the influence of noise enhancement at the receivers by placing the diagonal weighted filters at both transmitter side and receiver side with square root. Compared to previously proposed non-linear precoding methods including vector perturbation (VP), the proposed THP achieves high BER performance. Furthermore, we show that the proposed THP method is implemented with lower computational complexity than that of existing modified THP and VP in literature.
Juinn-Horng DENG Shiang-Chyun JHAN Sheng-Yang HUANG
A precoding design for double space-time block coding (STBC) system is investigated in this paper, i.e., the joint processing of STBC and dirty paper coding (DPC) techniques. These techniques are used for avoiding dual spatial streams interference and improving the transmitter diversity. The DPC system is interference free on multi-user or multi-antenna. The STBC transceiver can provide the transmit diversity. Due to the benefits about offered by the STBC and DPC techniques, we propose a new scheme called STBC-DPC system. The transceiver design involves the following procedures. First, the ordering QR decomposition of channel matrix and the maximum likelihood (ML) one-dimensional searching algorithm are proposed to acquire reliable performance. Next, the channel on/off assignment using the water filling algorithm, i.e., maximum capacity criterion, is proposed to overcome the deep fading channel problem. Finally, the STBC-DPC system with the modulus operation to limit the transmit signal level, i.e., the Tomlinson-Harashima precoding (THP) scheme, is proposed to achieve low peak-to-average power ratio (PAPR) performance. Simulation results confirm that the proposed STBC-DPC/THP with water filling ML algorithm can provide the low PAPR and excellent bit error rate (BER) performances.
Young-Sik EOM Jong Wook KWAK Seong-Tae JHANG Chu-Shik JHON
Chip Multiprocessors (CMPs) allow different applications to share LLC (Last Level Cache). Since each application has different cache capacity demand, LLC capacity should be partitioned in accordance with the demands. Existing partitioning algorithms estimate the capacity demand of each core by stack processing considering the LRU (Least Recently Used) replacement policy only. However, anti-thrashing replacement algorithms like BIP (Binary Insertion Policy) and BIP-Bypass emerged to overcome the thrashing problem of LRU replacement policy in a working set greater than the available cache size. Since existing stack processing cannot estimate the capacity demand with anti-thrashing replacement policy, partitioning algorithms also cannot partition cache space with anti-thrashing replacement policy. In this letter, we prove that BIP replacement policy is not feasible to stack processing but BIP-bypass is. We modify stack processing to accommodate BIP-Bypass. In addition, we propose the pipelined hardware of modified stack processing. With this hardware, we can get the success function of the various capacities with anti-thrashing replacement policy and assess the cache capacity of shared cache adequate to each core in real time.
Wenbing JIN Xuanya LI Yanyong YU Yongzhi WANG
To improve Last-Level Cache (LLC) management, numerous approaches have been proposed requiring additional hardware budget and increased overhead. A number of these approaches even change the organization of the existing cache design. In this letter, we propose Adaptive Insertion and Promotion (AIP) policies based on Least Recently Used (LRU) replacement. AIP dynamically inserts a missed line in the middle of the cache list and promotes a reused line several steps left, realizing the combination of LRU and LFU policies deliberately under a single unified scheme. As a result, it benefits workloads with high locality as well as with many frequently reused lines. Most importantly, AIP requires no additional hardware other than a typical LRU list, thus it can be easily integrated into the existing hardware with minimal changes. Other issues around LLC such as scans, thrashing and dead lines are all explored in our study. Experimental results on the gem5 simulator with SPEC CUP2006 benchmarks indicate that AIP outperforms LRU replacement policy by an average of 5.8% on the misses per thousand instructions metric.
Juinn-Horng DENG Sheng-Yang HUANG
The single carrier block transmission (SCBT) system has become one of the most popular modulation systems because of its low peak to average power ratio (PAPR). This work proposes precoding design on the transmitter side to retain low PAPR, improve performance, and reduce computational complexity on the receiver side. The system is designed according to the following procedure. First, upper-triangular dirty paper coding (UDPC) is utilized to pre-cancel the interference among multiple streams and provide a one-tap time-domain equalizer for the SCBT system. Next, to solve the problem of the high PAPR of the UDPC precoding system, Tomlinson-Harashima precoding (THP) is developed. Finally, since the UDPC-THP system is degraded by the deep fading channels, the dynamic channel on/off assignment by the maximum capacity algorithm (MCA) and minimum BER algorithm (MBA) is proposed to enhance the bit error rate (BER) performance. Simulation results reveal that the proposed precoding transceiver can provide excellent BER and low PAPR performances for the SCBT system over a multipath fading channel.
This paper proposes a downlink multi-user transmission scheme for the amplify-and-forward(AF)-based multi-relay cellular network, in which Tomlinson-Harashima precoding(TH precoding) and interference alignment(IA) are jointly applied. The whole process of transmission is divided into two phases: TH precoding is first performed at base-station(BS) to support the multiplexing of data streams transmitted to both mobile-stations(MS) and relay-stations(RS), and then IA is performed at both BS and RSs to achieve the interference-free communication. During the whole process, neither data exchange nor strict synchronization is required among BS and RSs thus reducing the cooperative complexity as well as improving the system performance. Theoretical analysis is provided with respect to the channel capacity of different types of users, resulting the upper-bounds of channel capacity. Our analysis and simulation results show that the joint applications of TH precoding and IA outperforms other schemes in the presented multi-relay cellular network.
Xun-yong Zhang Chen HE Ling-ge JIANG
In this paper, we propose a successive signal-to-leakage-plus-noise ratio (SLNR) based precoding with geometric mean decomposition (GMD) for the downlink multi-user multiple-input multiple-output (MU-MIMO) systems. The known leakages are canceled at the transmit side, and SLNR is calculated with the unknown leakages. GMD is applied to cancel the known leakages, so the subchannels for each receiver have equal gain. We further improve the proposed precoding scheme by ordering users. Simulation results show that the proposed schemes have a considerable bit error rate (BER) improvement over the original SLNR scheme.
Yanzhi SUN Muqing WU Jianming LIU Chaoyi ZHANG
In this letter, a quantization error-aware Tomlinson-Harashinma Precoding (THP) is proposed based on the equivalent zero-forcing (ZF) criterion in Multiuser Multiple-Input Single-Output (MU-MISO) systems with limited feedback, where the transmitter has only quantized channel direction information (CDI). This precoding scheme is robust to the channel uncertainties arising from the quantization error and the lack of channel magnitude information (CMI). Our simulation results show that the new THP scheme outperforms the conventional precoding scheme in limited feedback systems with respect to Bit Error Ratio (BER).
Takaaki KOGA Toru MATSUURA Sébastien FANIEL Satofumi SOUMA Shunsuke MINESHIGE Yoshiaki SEKINE Hiroki SUGIYAMA
We recently determined the values of intrinsic spin-orbit (SO) parameters for In0.52Al0.48As/In0.53Ga0.47As(10 nm)/In0.52Al0.48As (InGaAs/InAlAs) quantum wells (QW), lattice-matched to (001) InP, from the weak localization/antilocalization analysis of the low-temperature magneto-conductivity measurements [1]. We have then studied the subband energy spectra for the InGaAs/InAlAs double QW system from beatings in the Shubnikov de Haas (SdH) oscillations. The basic properties obtained here for the double QW system provides useful information for realizing nonmagnetic spin-filter devices based on the spin-orbit interaction [2].
Masanari FUJITA Mitsufumi SAITO Michihiko SUHARA
In this paper, we analyze current-voltage characteristics of InSb/AlInSb triple-barrier resonant tunneling diodes (TBRTDs) with spin-splitting under zero magnetic fields. The InSb has very small effective mass, thus we can obtain large spin-splitting by Rashba spin-orbit interaction due to asymmetric InSb/AlInSb quantum wells. In our model, broadening of each resonant tunneling level and spin-splitting energy can be considered to calculate spin-polarized resonant tunneling current.
Bing HUI Manar MOHAISEN KyungHi CHANG
Tomlinson-Harashima precoding (THP) is considered to be a prominent precoding scheme due to its ability to efficiently cancel out the known interference at the transmitter side. Therefore, the information rates achieved by THP are superior to those achieved by conventional linear precoding schemes. In this paper, new lower bounds on the achievable information rates for the regularized THP scheme are derived. Analytical results show that the lower bounds derived in this paper are tighter than the original lower bounds particularly for the low SNR range, while all lower bounds converge to as SNR ∞.
Pengyu GUAN Hans Christian Hansen MULVAD Yutaro TOMIYAMA Toshiyuki HIRANO Toshihiko HIROOKA Masataka NAKAZAWA
We demonstrate a single-channel 1.28 Tbit/s-525 km transmission using OTDM of subpicosecond DQPSK signals. In order to cope with transmission impairments due to time-varying higher-order PMD, which is one of the major limiting factors in such a long-haul ultrahigh-speed transmission, we newly developed an ultrafast time-domain optical Fourier transformation technique in a round-trip configuration. By applying this technique to subpicosecond pulses, transmission impairments were greatly reduced, and BER performance below FEC limit was obtained with increased system margin.