We propose an optimization method for the automatic design of approximate multipliers, which minimizes the average error according to the operand distributions. Our multiplier achieves up to 50.24% higher accuracy than the best reproduced approximate multiplier in DNNs, with 15.76% smaller area, 25.05% less power consumption, and 3.50% shorter delay. Compared with an exact multiplier, our multiplier reduces the area, power consumption, and delay by 44.94%, 47.63%, and 16.78%, respectively, with negligible accuracy losses. The tested DNN accelerator modules with our multiplier obtain up to 18.70% smaller area and 9.99% less power consumption than the original modules.

Approximate computing is an efficient approach to reduce the design complexity for error-resilient applications. Multipliers are key arithmetic units in many applications, such as deep neural networks (DNNs) and digital signal processing (DSP) systems. In this article, an open-source adaptable approximate multiplier design driven by input distribution and polarity is proposed to generate optimized approximate multipliers to trade off between the application-level performance and the hardware cost. The proposed method minimizes the average square of the absolute error of an approximate multiplier according to the probability distributions of operands extracted from the target application with consideration of input polarity, achieving low hardware cost and negligible application-level performance loss. The proposed method can generate unsigned multipliers (or signed multipliers) based on the Braun multiplier (or Baugh–Wooley multiplier). To demonstrate the effectiveness of the method, three different-scale quantized DNNs, including LeNet, AlexNet, and VGG16 with 8$\times $8 unsigned multiplication and an adaptive least mean square (LMS)-based finite impulse response (FIR) filter with 16$\times $16 fixed-point signed multiplication, are evaluated. In the DNN training process, a noise training technique is adopted to reduce the accuracy loss due to the approximation. When compared to the state-of-the-art approximate multipliers, the generated multipliers can achieve up to 26.4% and 27.1% product of power, delay, and area gains with negligible application-level performance loss in VGG16 and FIR applications, respectively.

Nonlinear activation functions (NAFs) play an essential role in deep neural networks (DNNs). Since versatile DNN accelerators need to support various DNNs which contain different NAFs, the flexible hardware design supporting those NAFs has become crucial. However, there are few high-precision flexible hardware architectures, and the symmetries of different NAFs have not been fully studied. This paper proposes a high-precision symmetry-aware architecture based on piecewise linear approximation. Through the reconfigurable data path, the architecture can support various typical NAFs. The efficient non-uniform segmentation scheme is proposed to achieve high precision for each NAF. Besides, the utilization of unified symmetry for NAFs can save half the memory. To reduce the computational cost, a 25×18 DSP is shared by two INT 7×9 multipliers with two independent inputs. The architecture is implemented on Xilinx ZC706 at a frequency of 410MHz. Compared with the state-of-the-art flexible nonlinear core, our flexible architecture costs fewer hardware resources with higher precision. Applying the design to BERT-BASE, MobileNetV3, and EfficientNet-B3 on the PyTorch platform, experimental results show that the accuracy loss is either 0 for BERT-BASE, or 0.002% for EfficientNet-B3. For MobileNetV3, the accuracy is even improved by 0.01%.

Lightweight convolutional neural networks (CNNs) have become increasingly popular due to their lower computational complexity and fewer memory accesses with equivalent accuracy compared to previous CNN models. However, the newly proposed networks bring new challenges to efficient hardware design, such as, in EfficientNet, depthwise convolution, squeeze-and-excitation (SE) module, and swish/sigmoid functions. Although individual engine architecture could achieve a high computing efficiency for the standard convolution or the depth-wise convolution, it is still not efficient for EfficientNet because the workload imbalance between two types of convolutional engines causes inevitable idling. To overcome this problem, we present a lightweight reconfigurable computational kernel based on FPGA with an exchangeable-track datapath scheme. In addition, a low-accuracy-loss function replacement strategy is proposed for swish/sigmoid functions. Furthermore, the low-cost hardware architecture to implement the replaced functions is designed. The proposed accelerator (LETA) can implement EfficientNet on Xilinx XCVU37P with a 300 MHz system clock and a 600 MHz kernel clock. The linear growth of resource usage in the 4-kernel implementation in 1 super logic region (SLR) with the same clock frequencies justifies the scalability of LETA. The experimental results show that LETA can achieve 2× throughput/DSP compared to the latest FPGA-based accelerator with 1.6% (0.7%) top-1 (top-5) accuracy loss on EfficientNet-B3.

We describe optimal cue mapping (OCM), a potentially eal-time binaural signal processing method for segregating sound source in the presence of multiple interfering 3D ound sources. Spatial cues are extracted from a multisource inaural mixture and used to train artificial neural etworks (ANNs) to estimate the spectral energy fraction of wanted speech source in the mixture. Once trained, the NN outputs form a spectral ratio mask which is applied rame-by-frame to the mixture to approximate the agnitude spectrum of the wanted speech. The speech ntelligibility performance of the OCM algorithm for nechoic sound sources is evaluated on previously unseen peech mixtures using the STOI automated measures, and ompared with an established reference method. The ptimized integration of multiple cues offers clear erformance benefits and the ability to quantify the relative mportance of each cue will facilitate computationally fficient implementations.

In this work, a genetic algorithm (GA) based frequency domain equalization (FDE) scheme referred to as FDE-GA, which does not require any guard interval (GI),is proposed for direct sequence-ultra wideband (DS-UWB) wireless communication systems and is shown to significantly outperform the RAKE receiver. The proposed FDE-GA receiver also has a dramatic complexity reduction over the previous RAKE-GA receiver, while achieving a comparable bit error rate (BER) performance. The FDE-GA structure achieves a much higher bandwidth efficiency than conventional FDE methods, because the inter-block-interference (IBI), as a result of the absence of the GI, is removed effectively within each block before the GA.

This study investigates Kalman smoothing (KS)-based frequency-domain channel estimation for uplink multiple-input multiple-output (MIMO) orthogonal frequency division multiple access (OFDMA) systems with time-varying channels. The proposed KS channel estimation scheme significantly outperforms the recursive least squares (RLS) channel estimation in the high signal-to-noise ratio (SNR) range, because of more effective exploitation of the signal information. In addition, channel interpolation is employed to improve the channel estimation accuracy by exploiting the correlation between adjacent subcarriers. The proposed KS channel estimator can also achieve a bit error rate (BER) performance which is close to the case with perfect channel state information (CSI) with a training overhead of only 5%.

This paper investigates carrier frequency offset (CFO) estimation and inphase/quadrature (I/Q) imbalance compensation in time-varying frequency-selective channels. We first propose a linear least squares (LLS) CFO estimation approach which has a lower complexity and a higher accuracy than the previous nonlinear CFO estimation methods. We then propose a Kalman filtering based I/Q imbalance compensation approach in the presence of CFO, which demonstrates a good ability to track the channel time variations with a fast convergence speed, by nulling the cyclic prefix (CP) and including the CFO in the state vector of the equivalent channel model. The proposed Kalman filtering based I/Q imbalance compensation approach with associated equalization tracks the time variation with a fast convergence speed. Simulation results show that the proposed compensation approach for CFO and I/Q imbalance provides a bit error rate (BER) performance close to the ideal case with perfect channel state information (CSI), no CFO and no I/Q imbalance.

We propose a novel semi-blind compensation scheme for both frequency-dependent and frequencyindependent I/Q imbalance based on independent component analysis (ICA) in multiple-input multiple-output (MIMO) orthogonal frequency division multiplexing (OFDM) systems, where ICA is applied to compensate for I/Q imbalance and equalize the received signal jointly, without any spectral overhead. A reference signal is embedded in the transmitted signal with little power consumption and no spectral overhead introduced, to enable ambiguity elimination for the ICA output signal at the receiver. Moreover, channel interpolation is incorporated with layered space frequency equalization (LSFE) to enhance the system performance. Simulation results show that the proposed implicit compensation scheme can not only provide a better bit error rate (BER) performance and a higher bandwidth efficiency than the previous training based I/Q imbalance compensation method, but also outperform the ideal case with perfect channel state information (CSI) and no I/Q imbalance, due to additional frequency diversity.

We propose an optimal asymmetric resource allocation scheme with limited feedback for wireless relay systems based on orthogonal frequency division multiplexing (OFDM) modulation. Unlike the previous work, we assume that the bits transmitted over the same subchannel from the source to the relay may be distributed over different subchannels when being forwarded from the relay to the destination. This not only increases the degree of freedom for transmission, but also reduces the transmission overhead significantly because the amount of feedback information required is independent of the number of subchannels. Furthermore, the transmission durations at the source and the relay are designed to be asymmetric, which also enhances the degree of freedom for transmission. Simulation results show that the proposed optimal asymmetric resource allocation scheme achieves a higher system capacity, and has less sensitivity to channel estimation errors than the previous work. It also demonstrates a fast convergence speed.

I/Q imbalance and carrier frequency offset (CFO) are two typical radio frequency (RF) circuit analog impairments in wireless communication systems, and degrade the system performance severely. In this paper, we propose a novel Kalman filtering based compensation scheme for I/Q imbalance and CFO in time-varying MIMO OFDM systems. To circumvent CFO and track time variations of wireless communication channels, the CFO is absorbed into the state vector of the equivalent channel model. Moreover, the inter-carrier interference (ICI) caused by phase shift of the cyclic prefix (CP) in the equivalent system is removed by decision feedback filtering, which allows low complexity compensation on each subcarrier independently. Simulation results show that the proposed approach can compensate for I/Q imbalance and CFO effectively and is robust again time variations.

We propose a dynamic resource allocation scheme for a wireless cooperative network based on orthogonal frequency division multiplexing (OFDM) modulations with limited feedback. The amount of the feedback information required from the relay to the source in the proposed scheme is independent of the number of subchannels, which results in a dramatic reduced overhead over conventional resource allocation schemes. Furthermore, the transmission durations at the source and the relay are designed to be asymmetric, which enhances the degree of freedom for transmission. Simulation results show that the proposed resource allocation scheme achieves a higher system capacity, and has less sensitivity to channel estimation errors than the previously proposed schemes based on subchannel paring and power allocation. It also has a fast convergence speed. An intensive performance analysis is provided for the proposed resource allocation scheme.

I/Q imbalance, which is one of the radio frequency (RF) circuit impairments in direct conversion transmitter and receiver, introduces severe performance degradation in wireless communication systems. In this paper, we propose a novel blind compensation algorithm for both frequency-dependent and frequency-independent I/Q imbalance based on independent component analysis (ICA) in multiple input multiple output (MIMO) orthogonal frequency division multiplexing (OFDM) systems, where ICA, an efficient higher order statistics (HOS) based blind source separation technique, is applied to compensate for I/Q imbalance and equalize the received signals simultaneously. Moreover, preceding is employed to resolve the ambiguity in the ICA output signals. Simulation results show that the proposed approach can not only compensate for I/Q imbalance effectively, but also achieve frequency diversity gains and outperform the case with perfect channel state information (CSI) and no I/Q imbalance.

We propose a genetic algorithm (GA) based equalization approach for direct sequence ultra-wideband (DS-UWB) wireless communications, where GA is combined with a RAKE receiver to combat the inter-symbol interference (ISI) due to the frequency selective nature of UWB channels for high data rate transmission. Simulation results show that the proposed GA based structure significantly outperforms the RAKE receiver. It also provides a close bit error rate (BER) performance to the optimal maximum likelihood detection (MLD) approach, while requiring a much lower computational complexity.

We propose a non-redundant linear precoding scheme and three peak-to-average power ratio (PAPR) reduction schemes for multiple-input multiple-output (MIMO) orthogonal frequency division multiplexing (OFDM) systems with independent component analysis (ICA) based blind equalization. The proposed precoding at the transmitter allows complete elimination of the ambiguity in the ICA equalized signals under certain conditions. The optimal design of precoding is investigated, and performance analysis on the ambiguity error probability is provided. The proposed PAPR reduction schemes are incorporated with precoding, and therefore do not introduce any spectral overhead compared to conventional PAPR reduction schemes. Simulation results show that the proposed blind structure provides a bit error rate (BER) performance which is much better than that of the subspace method, and close to the case with perfect channel state information (CSI) at the receiver. Furthermore, the proposed structure can reduce the PAPR of the transmit signals considerably.

A new stick text segmentation method based on sub connected area analysis is introduced in this paper. The foundation of this method is the sub connected area representation of a text image that can represent all connected areas in an image efficiently. This method consists mainly of four steps: sub connected area classification; finding the initial boundary following point; finding the optimal segmentation point by boundary tracing; and text segmentation. This method is similar to the boundary analysis method but is more efficient.

A method based on connected area detection is developed to segment and recognize dimension texts in engineering drawings. First we propose an effective algorithm to find all connected areas from drawing image. Then we use size criteria to find out character candidates from all connected areas and use collinear criteria to group separate character candidates into text strings. Finally we analyze text strings according to text patterns summarized from dimension texts and give out the recognition result.